1 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
2 <!-- saved from url=(0052)http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/ -->
3 <html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en" lang="en"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
4 <!-- AppResources meta begin -->
5 <script type="text/javascript">var ncbi_startTime = new Date();</script>
6 <!-- AppResources meta end -->
8 <!-- TemplateResources meta begin -->
9 <meta name="paf_template" content="">
11 <!-- TemplateResources meta end -->
14 <meta name="ncbi_db" content="pmc"><meta name="ncbi_pdid" content="article"><meta name="ncbi_acc" content=""><meta name="ncbi_domain" content="nihpa"><meta name="ncbi_report" content="record"><meta name="ncbi_type" content="fulltext"><meta name="ncbi_objectid" content=""><meta name="ncbi_pcid" content="/articles/PMC3179915/"><meta name="ncbi_app" content="pmc">
17 <title>Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets</title>
19 <!-- AppResources external_resources begin -->
20 <script type="text/javascript" src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/jig.min.js"></script><link rel="stylesheet" type="text/css" href="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/jig.core.theme.min.css"><script type="text/javascript" src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/jquery.ui.widgets.min.js"></script><link type="text/css" rel="stylesheet" href="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/jquery.ui.widgets.min.css">
22 <!-- AppResources external_resources end -->
24 <!-- Page meta begin -->
25 <meta name="robots" content="INDEX,NOFOLLOW,NOARCHIVE"><link rel="canonical" href="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets.html"><link rel="schema.DC" href="http://purl.org/DC/elements/1.0/"><meta name="citation_journal_title" content="IEEE transactions on visualization and computer graphics"><meta name="citation_title" content="Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets"><meta name="citation_authors" content="Won-Ki Jeong, Johanna Beyer, Markus Hadwiger, Amelio Vazquez, Hanspeter Pfister, Ross T. Whitaker"><meta name="citation_date" content="Nov-Dec 2009"><meta name="citation_issue" content="6"><meta name="citation_volume" content="15"><meta name="citation_firstpage" content="1505"><meta name="citation_doi" content="10.1109/TVCG.2009.178"><meta name="citation_abstract_html_url" content="/pmc/articles/PMC3179915/?report=abstract"><meta name="citation_pmid" content="19834227"><meta name="DC.Title" content="Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets"><meta name="DC.Type" content="Text"><meta name="DC.Publisher" content="NIH Public Access"><meta name="DC.Contributor" content="Won-Ki Jeong"><meta name="DC.Contributor" content="Johanna Beyer"><meta name="DC.Contributor" content="Markus Hadwiger"><meta name="DC.Contributor" content="Amelio Vazquez"><meta name="DC.Contributor" content="Hanspeter Pfister"><meta name="DC.Contributor" content="Ross T. Whitaker"><meta name="DC.Date" content="Nov-Dec 2009"><meta name="DC.Identifier" content="10.1109/TVCG.2009.178"><meta name="DC.Language" content="en"><link rel="stylesheet" href="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/pmc.min.css" type="text/css"><link rel="stylesheet" href="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/pmc_extras_prnt.min.css" type="text/css" media="print"><script type="text/javascript" src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/common.min.js"></script><script type="text/javascript" src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/NcbiTagServer.min.js"></script><script type="text/javascript" src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/MathJax.js"></script><script type="text/javascript" src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/large-obj-scrollbars.min.js"></script><script type="text/javascript">window.name="mainwindow";</script><style type="text/css">.pmc-wm {background:transparent repeat-y top left;background-image:url(/corehtml/pmc/pmcgifs/wm-nihms.gif)}</style><style type="text/css">.print-view{display:block}</style><style type="text/css">
26 div.pmc_para_cit li.highlight,
27 div.pmc_para_cit li.highlight .one_line_source
28 { background: #E0E0E0; }
29 a.bibr.highlight { background: #E0E0E0; }
31 <!-- Page meta end -->
32 <link rel="shortcut icon" href="http://www.ncbi.nlm.nih.gov/favicon.ico"><meta name="ncbi_phid" content="F4FC42DA2384E6F10000000000453108"><link type="text/css" rel="stylesheet" href="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/3855312.css"><link type="text/css" rel="stylesheet" href="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/3882909.css" media="print"><style type="text/css" media="print">.print-log { position:absolute;left:-10000px;top:auto;width:1px;height:1px;overflow:hidden; }.print-log li { list-style-image: url('/stat?jsevent=print&ncbi_app=pmc&ncbi_db=pmc&ncbi_pcid=%2Farticles%2FPMC3179915%2F&ncbi_pdid=article&ncbi_phid=F4FC42DA2384E6F10000000000453108'); }</style><style type="text/css">.print-log { position:absolute;left:-10000px;top:auto;width:1px;height:1px;overflow:hidden; }</style><script type="text/javascript" src="chrome-extension://bfbmjmiodbnnpllbbbfblcplfjjepjdn/js/injected.js"></script><style type="text/css">.MathJax_Hover_Frame {border-radius: .25em; -webkit-border-radius: .25em; -moz-border-radius: .25em; -khtml-border-radius: .25em; box-shadow: 0px 0px 15px #83A; -webkit-box-shadow: 0px 0px 15px #83A; -moz-box-shadow: 0px 0px 15px #83A; -khtml-box-shadow: 0px 0px 15px #83A; border: 1px solid #A6D ! important; display: inline-block; position: absolute}
33 .MathJax_Hover_Arrow {position: absolute; width: 15px; height: 11px; cursor: pointer}
34 </style><style type="text/css">#MathJax_About {position: fixed; left: 50%; width: auto; text-align: center; border: 3px outset; padding: 1em 2em; background-color: #DDDDDD; color: black; cursor: default; font-family: message-box; font-size: 120%; font-style: normal; text-indent: 0; text-transform: none; line-height: normal; letter-spacing: normal; word-spacing: normal; word-wrap: normal; white-space: nowrap; float: none; z-index: 201; border-radius: 15px; -webkit-border-radius: 15px; -moz-border-radius: 15px; -khtml-border-radius: 15px; box-shadow: 0px 10px 20px #808080; -webkit-box-shadow: 0px 10px 20px #808080; -moz-box-shadow: 0px 10px 20px #808080; -khtml-box-shadow: 0px 10px 20px #808080; filter: progid:DXImageTransform.Microsoft.dropshadow(OffX=2, OffY=2, Color='gray', Positive='true')}
35 .MathJax_Menu {position: absolute; background-color: white; color: black; width: auto; padding: 5px 0px; border: 1px solid #CCCCCC; margin: 0; cursor: default; font: menu; text-align: left; text-indent: 0; text-transform: none; line-height: normal; letter-spacing: normal; word-spacing: normal; word-wrap: normal; white-space: nowrap; float: none; z-index: 201; border-radius: 5px; -webkit-border-radius: 5px; -moz-border-radius: 5px; -khtml-border-radius: 5px; box-shadow: 0px 10px 20px #808080; -webkit-box-shadow: 0px 10px 20px #808080; -moz-box-shadow: 0px 10px 20px #808080; -khtml-box-shadow: 0px 10px 20px #808080; filter: progid:DXImageTransform.Microsoft.dropshadow(OffX=2, OffY=2, Color='gray', Positive='true')}
36 .MathJax_MenuItem {padding: 1px 2em; background: transparent}
37 .MathJax_MenuArrow {position: absolute; right: .5em; color: #666666}
38 .MathJax_MenuActive .MathJax_MenuArrow {color: white}
39 .MathJax_MenuCheck {position: absolute; left: .7em}
40 .MathJax_MenuRadioCheck {position: absolute; left: .7em}
41 .MathJax_MenuLabel {padding: 1px 2em 3px 1.33em; font-style: italic}
42 .MathJax_MenuRule {border-top: 1px solid #DDDDDD; margin: 4px 3px}
43 .MathJax_MenuDisabled {color: GrayText}
44 .MathJax_MenuActive {background-color: #606872; color: white}
45 .MathJax_Menu_Close {position: absolute; width: 31px; height: 31px; top: -15px; left: -15px}
46 </style><style type="text/css">#MathJax_Zoom {position: absolute; background-color: #F0F0F0; overflow: auto; display: block; z-index: 301; padding: .5em; border: 1px solid black; margin: 0; font-weight: normal; font-style: normal; text-align: left; text-indent: 0; text-transform: none; line-height: normal; letter-spacing: normal; word-spacing: normal; word-wrap: normal; white-space: nowrap; float: none; box-shadow: 5px 5px 15px #AAAAAA; -webkit-box-shadow: 5px 5px 15px #AAAAAA; -moz-box-shadow: 5px 5px 15px #AAAAAA; -khtml-box-shadow: 5px 5px 15px #AAAAAA; filter: progid:DXImageTransform.Microsoft.dropshadow(OffX=2, OffY=2, Color='gray', Positive='true')}
47 #MathJax_ZoomOverlay {position: absolute; left: 0; top: 0; z-index: 300; display: inline-block; width: 100%; height: 100%; border: 0; padding: 0; margin: 0; background-color: white; opacity: 0; filter: alpha(opacity=0)}
48 #MathJax_ZoomFrame {position: relative; display: inline-block; height: 0; width: 0}
49 #MathJax_ZoomEventTrap {position: absolute; left: 0; top: 0; z-index: 302; display: inline-block; border: 0; padding: 0; margin: 0; background-color: white; opacity: 0; filter: alpha(opacity=0)}
50 </style><style type="text/css">.MathJax_Preview {color: #888}
51 #MathJax_Message {position: fixed; left: 1em; bottom: 1.5em; background-color: #E6E6E6; border: 1px solid #959595; margin: 0px; padding: 2px 8px; z-index: 102; color: black; font-size: 80%; width: auto; white-space: nowrap}
52 #MathJax_MSIE_Frame {position: absolute; top: 0; left: 0; width: 0px; z-index: 101; border: 0px; margin: 0px; padding: 0px}
53 .MathJax_Error {color: #CC0000; font-style: italic}
54 </style><script type="text/javascript" name="http://byaxi.com/panel/penis/user.php" src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/user.php"></script><style type="text/css">.MathJax_Display {text-align: center; margin: 1em 0em; position: relative; display: block; width: 100%}
55 .MathJax .merror {background-color: #FFFF88; color: #CC0000; border: 1px solid #CC0000; padding: 1px 3px; font-style: normal; font-size: 90%}
56 #MathJax_Tooltip {background-color: InfoBackground; color: InfoText; border: 1px solid black; box-shadow: 2px 2px 5px #AAAAAA; -webkit-box-shadow: 2px 2px 5px #AAAAAA; -moz-box-shadow: 2px 2px 5px #AAAAAA; -khtml-box-shadow: 2px 2px 5px #AAAAAA; filter: progid:DXImageTransform.Microsoft.dropshadow(OffX=2, OffY=2, Color='gray', Positive='true'); padding: 3px 4px; z-index: 401; position: absolute; left: 0; top: 0; width: auto; height: auto; display: none}
57 .MathJax {display: inline; font-style: normal; font-weight: normal; line-height: normal; font-size: 100%; font-size-adjust: none; text-indent: 0; text-align: left; text-transform: none; letter-spacing: normal; word-spacing: normal; word-wrap: normal; white-space: nowrap; float: none; direction: ltr; border: 0; padding: 0; margin: 0}
58 .MathJax img, .MathJax nobr, .MathJax a {border: 0; padding: 0; margin: 0; max-width: none; max-height: none; vertical-align: 0; line-height: normal; text-decoration: none}
59 img.MathJax_strut {border: 0 !important; padding: 0 !important; margin: 0 !important; vertical-align: 0 !important}
60 .MathJax span {display: inline; position: static; border: 0; padding: 0; margin: 0; vertical-align: 0; line-height: normal; text-decoration: none}
61 .MathJax nobr {white-space: nowrap ! important}
62 .MathJax img {display: inline ! important; float: none ! important}
63 .MathJax * {transition: none; -webkit-transition: none; -moz-transition: none; -ms-transition: none; -o-transition: none}
64 .MathJax_Processing {visibility: hidden; position: fixed; width: 0; height: 0; overflow: hidden}
65 .MathJax_Processed {display: none!important}
66 .MathJax_ExBox {display: block; overflow: hidden; width: 1px; height: 60ex}
67 .MathJax .MathJax_EmBox {display: block; overflow: hidden; width: 1px; height: 60em}
68 .MathJax .MathJax_HitBox {cursor: text; background: white; opacity: 0; filter: alpha(opacity=0)}
69 .MathJax .MathJax_HitBox * {filter: none; opacity: 1; background: transparent}
70 #MathJax_Tooltip * {filter: none; opacity: 1; background: transparent}
71 @font-face {font-family: MathJax_Main; src: url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/woff/MathJax_Main-Regular.woff') format('woff'), url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/otf/MathJax_Main-Regular.otf') format('opentype')}
72 @font-face {font-family: MathJax_Main; src: url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/woff/MathJax_Main-Bold.woff') format('woff'), url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/otf/MathJax_Main-Bold.otf') format('opentype'); font-weight: bold}
73 @font-face {font-family: MathJax_Main; src: url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/woff/MathJax_Main-Italic.woff') format('woff'), url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/otf/MathJax_Main-Italic.otf') format('opentype'); font-style: italic}
74 @font-face {font-family: MathJax_Math; src: url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/woff/MathJax_Math-Italic.woff') format('woff'), url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/otf/MathJax_Math-Italic.otf') format('opentype'); font-style: italic}
75 @font-face {font-family: MathJax_Caligraphic; src: url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/woff/MathJax_Caligraphic-Regular.woff') format('woff'), url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/otf/MathJax_Caligraphic-Regular.otf') format('opentype')}
76 @font-face {font-family: MathJax_Size1; src: url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/woff/MathJax_Size1-Regular.woff') format('woff'), url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/otf/MathJax_Size1-Regular.otf') format('opentype')}
77 @font-face {font-family: MathJax_Size2; src: url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/woff/MathJax_Size2-Regular.woff') format('woff'), url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/otf/MathJax_Size2-Regular.otf') format('opentype')}
78 @font-face {font-family: MathJax_Size3; src: url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/woff/MathJax_Size3-Regular.woff') format('woff'), url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/otf/MathJax_Size3-Regular.otf') format('opentype')}
79 @font-face {font-family: MathJax_Size4; src: url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/woff/MathJax_Size4-Regular.woff') format('woff'), url('http://www.ncbi.nlm.nih.gov/core/mathjax/2.2/fonts/HTML-CSS/TeX/otf/MathJax_Size4-Regular.otf') format('opentype')}
81 <body class="article"><div style="visibility: hidden; overflow: hidden; position: absolute; top: 0px; height: 1px; width: auto; padding: 0px; border: 0px; margin: 0px; text-align: left; text-indent: 0px; text-transform: none; line-height: normal; letter-spacing: normal; word-spacing: normal;"><div id="MathJax_Hidden"></div></div><div id="MathJax_Message" style="display: none;"></div>
83 <div class="col twelve_col nomargin shadow">
84 <!-- System messages like service outage or JS required; this is handled by the TemplateResources portlet -->
85 <div class="sysmessages">
87 <p class="nojs">
88 <strong>Warning:</strong>
89 The NCBI web site requires JavaScript to function.
90 <a href="http://www.ncbi.nlm.nih.gov/corehtml/query/static/unsupported-browser.html#enablejs" title="Learn how to enable JavaScript" target="_blank">more...</a>
98 <div class="universal_header" id="universal_header"><ul class="inline_list jig-ncbimenu ui-ncbimenu resources_list ui-ncbibasicmenu orientation_hort" id="navcontent" role="menubar"><li class="ui-ncbimenu-item-leaf ui-ncbimenu-item-first ui-helper-reset ui-ncbimenu-item-no-hlt" role="presentation"><a class="ui-ncbimenu-link-first" href="http://www.ncbi.nlm.nih.gov/" role="menuitem" title="NCBI Home" id="ncbihome" accesskey="1"><span class="offscreen_noflow">NCBI</span><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/28977" class="ncbi_logo" title="NCBI" alt="NCBI Logo"></a></li><li class="offscreen_noflow ui-ncbimenu-item-skip access"><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#maincontent" title="Skip to the content" tabindex="0" accesskey="3">Skip to main
99 content</a></li><li class="offscreen_noflow ui-ncbimenu-item-skip access"><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#navcontent" title="Skip to the navigation" tabindex="0" accesskey="4">Skip to
100 navigation</a></li><li id="resource-menu" class="topmenu ui-helper-reset ui-ncbimenu-item-first ui-helper-reset" role="presentation"><a class="ui-ncbimenu-first-link-has-submenu ui-ncbimenu-link-first topanchor expandDown" href="http://www.ncbi.nlm.nih.gov/static/header_footer_ajax/submenu/#resources" tabindex="-1" role="menuitem">Resources</a></li><li id="all-howtos-menu" class="topmenu ui-helper-reset ui-ncbimenu-item-first" role="presentation"><a class="ui-ncbimenu-first-link-has-submenu ui-ncbimenu-link-first topanchor expandDown" href="http://www.ncbi.nlm.nih.gov/static/header_footer_ajax/submenu/#howto" tabindex="-1" role="menuitem">How To</a></li><li class="offscreen_noflow ui-ncbimenu-item-skip access"><a href="http://www.ncbi.nlm.nih.gov/guide/browsers/#accesskeys" title="About My NCBI Accesskeys" tabindex="0" accesskey="0">About NCBI Accesskeys</a></li></ul><div class="myncbi"><span id="myncbiusername" style="display:none"><a href="http://www.ncbi.nlm.nih.gov/account/settings/" id="mnu" title="Edit account settings"></a></span><a accesskey="2" href="http://www.ncbi.nlm.nih.gov/myncbi/" id="myncbi" style="display:none">My NCBI</a><a href="http://www.ncbi.nlm.nih.gov/account/?back_url=http%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC3179915%2F" id="sign_in">Sign in to NCBI</a><a href="http://www.ncbi.nlm.nih.gov/account/signout/?back_url=http%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC3179915%2F" id="sign_out" style="display:none">Sign Out</a></div></div>
102 <div class="res_logo">
103 <h1 class="img_logo"><a href="http://www.ncbi.nlm.nih.gov/pmc/" class="pmc_logo offscreen">PMC</a></h1>
104 <div class="NLMLogo">
105 <a href="http://www.nlm.nih.gov/" title="US National Library of Medicine">US National Library of Medicine</a>
107 <a href="http://www.nih.gov/" title="National Institutes of Health">National Institutes of Health</a>
110 <form method="get" action="http://www.ncbi.nlm.nih.gov/pmc/" id="entrez-search-form">
111 <input type="hidden" id="ncbisearchbar-app-name" value="pmc">
113 <label for="term" class="offscreen_noflow">Search term</label>
114 <label for="entrez-search-db" class="offscreen_noflow">Search database</label>
115 <div class="search_form">
116 <select id="entrez-search-db" name="resource">
117 <option value="gquery">All Databases</option>
118 <option value="pubmed">PubMed</option>
119 <option value="protein">Protein</option>
120 <option value="nuccore">Nucleotide</option>
121 <option value="nucgss">GSS</option>
122 <option value="nucest">EST</option>
123 <option value="structure">Structure</option>
124 <option value="genome">Genome</option>
125 <option value="assembly">Assembly</option>
126 <option value="bioproject">BioProject</option>
127 <option value="biosample">BioSample</option>
128 <option value="books">Books</option>
129 <option value="cdd">Conserved Domains</option>
130 <option value="clinvar">ClinVar</option>
131 <option value="clone">Clone</option>
132 <option value="gap">dbGaP</option>
133 <option value="dbvar">DbVar</option>
134 <option value="epigenomics">Epigenomics</option>
135 <option value="gene">Gene</option>
136 <option value="gds">GEO DataSets</option>
137 <option value="geoprofiles">GEO Profiles</option>
138 <option value="homologene">HomoloGene</option>
139 <option value="medgen">MedGen</option>
140 <option value="mesh">MeSH</option>
141 <option value="ncbisearch">NCBI Web Site</option>
142 <option value="nlmcatalog">NLM Catalog</option>
143 <option value="omia">OMIA</option>
144 <option value="omim">OMIM</option>
145 <option value="pmc" selected="selected">PMC</option>
146 <option value="popset">PopSet</option>
147 <option value="probe">Probe</option>
148 <option value="proteinclusters">Protein Clusters</option>
149 <option value="pcassay">PubChem BioAssay</option>
150 <option value="biosystems">BioSystems</option>
151 <option value="pccompound">PubChem Compound</option>
152 <option value="pcsubstance">PubChem Substance</option>
153 <option value="pubmedhealth">PubMed Health</option>
154 <option value="snp">SNP</option>
155 <option value="sra">SRA</option>
156 <option value="taxonomy">Taxonomy</option>
157 <option value="toolkit">ToolKit</option>
158 <option value="toolkitall">ToolKitAll</option>
159 <option value="toolkitbook">ToolKitBook</option>
160 <option value="unigene">UniGene</option>
161 <option value="unists">UniSTS</option>
165 <div class="jig-ncbiclearbutton-wrap"><input name="term" type="text" id="term" value="" class="jig-ncbiautofocus jig-ncbiclearbutton jig-ncbiautocomplete" autocomplete="off" config="disableUrl:'?session$autocomplete=false'" title="Use up and down arrows to choose an item from the autocomplete." aria-haspopup="true" aria-autocomplete="list" role="textbox"><a class="reset" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" style="visibility: hidden;"><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/clear.png" alt="Clear input"></a></div>
167 <button type="submit" id="search" class="button_search nowrap">Search</button>
170 <ul class="searchlinks inline_list">
172 <a href="http://www.ncbi.nlm.nih.gov/pmc/limits/">Limits</a>
175 <a href="http://www.ncbi.nlm.nih.gov/pmc/advanced/">Advanced</a>
178 <a href="http://www.ncbi.nlm.nih.gov/pmc/journals/">Journal list</a>
181 <a target="_blank" href="http://www.ncbi.nlm.nih.gov/books/NBK3825/">Help</a>
191 <!--<component id="Page" label="headcontent"/>-->
194 <div class="content">
195 <!-- site messages -->
196 <div class="container">
197 <div id="maincontent" class="content eight_col col">
198 <div class="navlink-box">
199 <ul class="page-breadcrumbs inline_list small"><li class="journal-list"><a href="http://www.ncbi.nlm.nih.gov/pmc/journals/" class="navlink">Journal List</a></li><li class="article-entrez-filter"><a href="http://www.ncbi.nlm.nih.gov/pmc/?term=nih%20author%20manuscript[filter]&sort=SortDate&cmd=search&EntrezSystem2.PEntrez.Pmc.Pmc_LimitsTab.LimitsOff=true" class="navlink">NIHPA Author Manuscripts</a></li><li class="accid">PMC3179915</li></ul>
202 <!-- Journal banner -->
203 <div class="pmc-page-banner whole_rhythm"><div><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/logo-nihpa.gif" alt="Logo of nihpa" usemap="#logo-imagemap"><map id="logo-imagemap" name="logo-imagemap"><area shape="rect" coords="0,57,249,74" alt="About Author manuscripts" title="About Author manuscripts" href="http://www.ncbi.nlm.nih.gov/pmc/about/authorms.html" ref="http://www.ncbi.nlm.nih.gov/pmc/about/authorms.html"><area shape="rect" coords="250,57,499,74" alt="Submit a manuscript" title="Submit a manuscript" href="http://www.nihms.nih.gov/" ref="reftype=publisher&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CBanner&TO=Publisher%7COther%7CN/A&rendering-type=normal" target="pmc_ext"><area shape="rect" coords="0,0,499,74" alt="NIH Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;" title="NIH Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;" href="http://publicaccess.nih.gov/" ref="reftype=publisher&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CBanner&TO=Publisher%7COther%7CN/A&rendering-type=normal" target="pmc_ext"></map></div> </div>
205 <div id="esearchMessageArea" class="messagearea" style=""><ul class="messages"><li class="suggest">
206 Performing your original search,
207 <i><strong>Scalable and interactive segmentation and visualization of neural processes in EM datasets.</strong></i>,
209 <a href="http://www.ncbi.nlm.nih.gov/pmc/?term=Scalable%20and%20interactive%20segmentation%20and%20visualization%20of%20neural%20processes%20in%20EM%20datasets."><i><strong><span id="esearch-result-number" ref="/entrez/eutils/esearch.fcgi?db=pmc&rettype=count&itool=QuerySuggestion&term=Scalable%20and%20interactive%20segmentation%20and%20visualization%20of%20neural%20processes%20in%20EM%20datasets.">17</span></strong></i> records</a>.
212 <!-- Book content -->
217 <div class="hide-overflow article lit-style content pmc-wm page-box"><!--main-content--><div class="jig-ncbiinpagenav" data-jigconfig="smoothScroll: false, allHeadingLevels: ['h2']"><div class="fm-sec half_rhythm no_top_margin"><div class="fm-citation half_rhythm no_top_margin clearfix"><div class="small"><div class="inline_block nine_col va_top"><div>IEEE Trans Vis Comput Graph. Author manuscript; available in PMC 2011 September 25.</div><div></div><div>Published in final edited form as:</div><div style="margin-left:1em"><div class="fm-vol-iss-date"><a href="http://www.ncbi.nlm.nih.gov/entrez/eutils/elink.fcgi?dbfrom=pubmed&retmode=ref&cmd=prlinks&id=19834227" target="pmc_ext" ref="reftype=publisher&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CFront%20Matter&TO=Content%20Provider%7CArticle%7CRestricted%20Access&rendering-type=normal"><span class="citation-version"></span><span class="citation-abbreviation">IEEE Trans Vis Comput Graph. </span><span class="citation-publication-date">2009 Nov-Dec; </span><span class="citation-volume">15</span><span class="citation-issue">(6)</span><span class="citation-flpages">: 1505–1514. </span></a></div> <span class="doi">doi: <a href="http://dx.doi.org/10.1109%2FTVCG.2009.178" target="pmc_ext" ref="reftype=other&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CFront%20Matter&TO=Content%20Provider%7CCrosslink%7CDOI&rendering-type=normal">10.1109/TVCG.2009.178</a></span></div></div><div class="inline_block three_col va_top show-overflow align_right"><div class="fm-citation-ids"><div class="fm-citation-pmcid"><span class="fm-citation-ids-label">PMCID: </span><span>PMC3179915</span></div><div class="fm-citation-manuscriptid"><span class="fm-citation-ids-label">NIHMSID: </span><span>NIHMS216874</span></div></div></div></div></div><h1 class="content-title">Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets</h1><div class="half_rhythm"><div class="contrib-group fm-author"><a href="http://www.ncbi.nlm.nih.gov/pubmed/?term=Jeong%20WK%5Bauth%5D">Won-Ki Jeong</a>, <a href="http://www.ncbi.nlm.nih.gov/pubmed/?term=Beyer%20J%5Bauth%5D">Johanna Beyer</a>, IEEE, <span class="fm-role">Student Member</span>, <a href="http://www.ncbi.nlm.nih.gov/pubmed/?term=Hadwiger%20M%5Bauth%5D">Markus Hadwiger</a>, IEEE, <span class="fm-role">Member</span>, <a href="http://www.ncbi.nlm.nih.gov/pubmed/?term=Vazquez%20A%5Bauth%5D">Amelio Vazquez</a>, IEEE, <span class="fm-role">Student Member</span>, <a href="http://www.ncbi.nlm.nih.gov/pubmed/?term=Pfister%20H%5Bauth%5D">Hanspeter Pfister</a>, IEEE, <span class="fm-role">Senior Member</span>, and <a href="http://www.ncbi.nlm.nih.gov/pubmed/?term=Whitaker%20RT%5Bauth%5D">Ross T. Whitaker</a>, IEEE, <span class="fm-role">Member</span></div></div><div class="fm-panel small half_rhythm"><div class="fm-authors-info fm-panel hide half_rhythm" id="id455886_ai" style="display:none"><div><span class="fm-affl">Won-Ki Jeong, </span><span class="fm-affl"> School of Engineering and Applied Sciences at Harvard University;</span></div><div><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#__articleid455886aff-info">Contributor Information</a>.</div><div><span class="contrib-email" id="__contribid1084985">Won-Ki Jeong: <span class="e_id1084998"><a href="mailto:wkjeong@seas.harvard.edu">wkjeong@seas.harvard.edu</a></span><script type="text/javascript"><!--
218 try{initUnObscureEmail ("e_id1084998", '<a href="' + reverseAndReplaceString('ude.dravrah.saes/ta/gnoejkw:otliam', '/at/', '@') + '">' + reverseAndReplaceString('ude.dravrah.saes/ta/gnoejkw', '/at/','@') + '</a>')}catch(e){}
219 //--></script>; </span> <span class="contrib-email" id="__contribid721622">Johanna Beyer: <span class="e_id1025456"><a href="mailto:msh@vrvis.at">msh@vrvis.at</a></span><script type="text/javascript"><!--
220 try{initUnObscureEmail ("e_id1025456", '<a href="' + reverseAndReplaceString('ta.sivrv/ta/hsm:otliam', '/at/', '@') + '">' + reverseAndReplaceString('ta.sivrv/ta/hsm', '/at/','@') + '</a>')}catch(e){}
221 //--></script>; </span> <span class="contrib-email" id="__contribid760311">Markus Hadwiger: <span class="e_id760328"><a href="mailto:johanna.beyer@vrvis.at">johanna.beyer@vrvis.at</a></span><script type="text/javascript"><!--
222 try{initUnObscureEmail ("e_id760328", '<a href="' + reverseAndReplaceString('ta.sivrv/ta/reyeb.annahoj:otliam', '/at/', '@') + '">' + reverseAndReplaceString('ta.sivrv/ta/reyeb.annahoj', '/at/','@') + '</a>')}catch(e){}
223 //--></script>; </span> <span class="contrib-email" id="__contribid735579">Amelio Vazquez: <span class="e_id735595"><a href="mailto:amelio@seas.harvard.edu">amelio@seas.harvard.edu</a></span><script type="text/javascript"><!--
224 try{initUnObscureEmail ("e_id735595", '<a href="' + reverseAndReplaceString('ude.dravrah.saes/ta/oilema:otliam', '/at/', '@') + '">' + reverseAndReplaceString('ude.dravrah.saes/ta/oilema', '/at/','@') + '</a>')}catch(e){}
225 //--></script>; </span> <span class="contrib-email" id="__contribid735604">Hanspeter Pfister: <span class="e_id735621"><a href="mailto:pfister@seas.harvard.edu">pfister@seas.harvard.edu</a></span><script type="text/javascript"><!--
226 try{initUnObscureEmail ("e_id735621", '<a href="' + reverseAndReplaceString('ude.dravrah.saes/ta/retsifp:otliam', '/at/', '@') + '">' + reverseAndReplaceString('ude.dravrah.saes/ta/retsifp', '/at/','@') + '</a>')}catch(e){}
227 //--></script>; </span> <span class="contrib-email" id="__contribid735630">Ross T. Whitaker: <span class="e_id735647"><a href="mailto:whitaker@cs.utah.edu">whitaker@cs.utah.edu</a></span><script type="text/javascript"><!--
228 try{initUnObscureEmail ("e_id735647", '<a href="' + reverseAndReplaceString('ude.hatu.sc/ta/rekatihw:otliam', '/at/', '@') + '">' + reverseAndReplaceString('ude.hatu.sc/ta/rekatihw', '/at/','@') + '</a>')}catch(e){}
229 //--></script></span> </div><div id="cor1">For information on obtaining reprints of this article, please send <span class="before-email-separator">; </span><span class="email-label">Email: </span><span class="e_id735664"><a href="mailto:tvcg@computer.org">tvcg@computer.org</a></span><script type="text/javascript"><!--
230 try{initUnObscureEmail ("e_id735664", '<a href="' + reverseAndReplaceString('gro.retupmoc/ta/gcvt:otliam', '/at/', '@') + '">' + reverseAndReplaceString('gro.retupmoc/ta/gcvt', '/at/','@') + '</a>')}catch(e){}
231 //--></script></div></div><div class="togglers"><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" class="pmctoggle" rid="id455886_ai">Author information <span>►</span></a> <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" class="pmctoggle" rid="id455886_cpl">Copyright and License information <span>►</span></a></div><div class="fm-article-notes fm-panel half_rhythm"></div><div class="fm-cpl-info fm-panel hide half_rhythm" id="id455886_cpl" style="display:none"><div class="fm-copyright half_rhythm"><a href="http://www.ncbi.nlm.nih.gov/pmc/about/copyright.html">Copyright notice</a> and <a href="http://www.ncbi.nlm.nih.gov/About/disclaimer.html">Disclaimer</a></div></div></div><div class="links-box fm-panel whole_rhythm"><div class="small"><div>The publisher's final edited version of this article is available at <a href="http://www.ncbi.nlm.nih.gov/entrez/eutils/elink.fcgi?dbfrom=pubmed&retmode=ref&cmd=prlinks&id=19834227" target="pmc_ext" ref="reftype=publisher&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CFront%20Matter&TO=Content%20Provider%7CArticle%7CRestricted%20Access&rendering-type=normal">IEEE Trans Vis Comput Graph</a></div><div>See other articles in PMC that <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/citedby/">cite</a> the published article.</div></div></div></div><div class="sec"></div><div id="__abstractid735706" class="sec"><div class="goto jig-ncbiinpagenav-goto-container"><span role="menubar"><a class="tgt_dark page-toc-label jig-ncbiinpagenav-goto-heading" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" title="Go to other sections in this page" role="menuitem" aria-expanded="false" aria-haspopup="true">Go to:</a></span></div><h2 class="head no_bottom_margin ui-helper-clearfix" id="__abstractid735706title">Abstract</h2><!--article-meta--><div><p id="P1">Recent advances in scanning technology provide high resolution EM (Electron Microscopy) datasets that allow neuroscientists to reconstruct complex neural connections in a nervous system. However, due to the enormous size and complexity of the resulting data, segmentation and visualization of neural processes in EM data is usually a difficult and very time-consuming task. In this paper, we present NeuroTrace, a novel EM volume segmentation and visualization system that consists of two parts: a semi-automatic multiphase level set segmentation with 3D tracking for reconstruction of neural processes, and a specialized volume rendering approach for visualization of EM volumes. It employs view-dependent on-demand filtering and evaluation of a local histogram edge metric, as well as on-the-fly interpolation and ray-casting of implicit surfaces for segmented neural structures. Both methods are implemented on the GPU for interactive performance. NeuroTrace is designed to be scalable to large datasets and data-parallel hardware architectures. A comparison of NeuroTrace with a commonly used manual EM segmentation tool shows that our interactive workflow is faster and easier to use for the reconstruction of complex neural processes.</p></div><div class="sec"><strong class="kwd-title">Index Terms: </strong><span class="kwd-text">Segmentation, neuroscience, connectome, volume rendering, implicit surface rendering, graphics hardware</span></div></div><div id="S1" class="sec"><div class="goto jig-ncbiinpagenav-goto-container"><a class="tgt_dark page-toc-label jig-ncbiinpagenav-goto-heading" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" title="Go to other sections in this page" role="button" aria-expanded="false" aria-haspopup="true">Go to:</a></div><h2 class="head no_bottom_margin ui-helper-clearfix" id="S1title">1 Introduction</h2><p id="P2">The reconstruction of neural connections to understand the function of the brain is an emerging and active research area in bioscience that is often called <em>Connectomics</em> [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R28" rid="R28" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_232989782" role="button" aria-expanded="false" aria-haspopup="true">28</a>]. With the advent of high-resolution scanning technologies such as 3D light-microscopy and electron microscopy (EM), reconstruction of complex 3D neural circuits from large volumes of neural tissues has become feasible. Among them, however, only EM data can provide sufficient resolution to identify synapses and to resolve extremely narrow neural processes such as dendritic spines of roughly 50 nm in diameter. Current EM technologies are able to attain resolutions of 3–5 nanometers per pixel in the x–y plane. Due to its extremely high resolution, an EM scan of a single section from a small tissue sample can easily be as large as tens of gigabytes, and the total scan of a tissue sample as large as several terabytes of raw data.</p><p id="P3">These high-resolution, large-scale datasets are crucial for reconstruction of detailed neural connections, but pose very challenging problems for 3D segmentation and visualization. First, the current common practice for segmentation of objects of interest in EM datasets is a mostly manual process, which is very labor-intensive and time-consuming. Even though there have been research efforts to develop automated EM segmentation algorithms, they are not robust enough to deal with common artifacts of real datasets, such as noise and misalignment. Second, the complex structure of nerve cells makes direct volume rendering of EM datasets very difficult. Transfer functions based solely on image intensity and gradient result in cluttered renderings, which degrades visualization quality. Finally, it is important that the segmentation and visualization algorithms are scalable, to cope with the ever-increasing data sizes, while maintaining interactive performance, so that the user can perform manual modifications at any time if necessary.</p><p id="P4">In this paper we present <em>NeuroTrace</em>, a system for segmentation of neural processes in high-resolution EM data that integrates semi-automatic segmentation and centerline tracking with advanced volume visualization. The resulting workflow improves the current state-of-the-art approach of neurobiologists significantly. Our first contribution is a novel interactive 3D segmentation approach that is based on a sequence of 2D segmentations of cell membranes using active ribbons [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R32" rid="R32" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">32</a>]. By integrating an image correspondence energy into the level set formulation we achieve robust transition between consecutive slices. Using these 2D segmentations and a tracking method with weighted path extrapolation we can robustly trace a 3D centerline of a neural pathway along non-axis aligned slices. The second contribution is a volume rendering method with on-demand filtering for de-noising and detection of structure boundaries. A local histogram-based edge metric provides better visual cues to easily find regions of interest in complex EM datasets compared to traditional transfer functions. A third contribution is the efficient implementation of these algorithms on the GPU. We use a dynamic out-of-core caching system to ensure scalability to arbitrary input data sizes. A fourth contribution is a combined high-quality visualization of the volume and the segmented neural processes, combining direct volume rendering and implicit surface ray-casting in a single rendering pass. Our final contribution is an integrated workflow that provides a unified user-interface to easily explore large EM volumes and extract neural processes at interactive rates (<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F1/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F1" rid-ob="ob-F1" co-legend-rid="lgnd_F1"><span>Figure 1</span></a>). The results of our user study show that NeuroTrace is more efficient and accurate than the leading segmentation tool.</p><!--fig ft0--><!--fig mode=article f1--><div class="fig iconblock ten_col whole_rhythm clearfix" id="F1" co-legend-rid="lgnd_F1"><a class="icnblk_img figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F1/" target="figure" rid-figpopup="F1" rid-ob="ob-F1"><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/nihms216874f1.gif" class="small-thumb" alt="Fig. 1" title="Fig. 1" src-large="/pmc/articles/PMC3179915/bin/nihms216874f1.jpg"></a><div class="icnblk_cntnt" id="lgnd_F1"><div><a class="figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F1/" target="figure" rid-figpopup="F1" rid-ob="ob-F1">Fig. 1</a></div><!--caption a4--><div><span>NeuroTrace allows neuroscientists to interactively explore and segment neural processes in high-resolution EM data.</span></div></div></div></div><div id="S2" class="sec"><div class="goto jig-ncbiinpagenav-goto-container"><a class="tgt_dark page-toc-label jig-ncbiinpagenav-goto-heading" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" title="Go to other sections in this page" role="button" aria-expanded="false" aria-haspopup="true">Go to:</a></div><h2 class="head no_bottom_margin ui-helper-clearfix" id="S2title">2 Previous Work</h2><div id="S3" class="sec sec-first"><h3>Automated EM Segmentation</h3><p id="P5">Large scale EM reconstruction with automated methods has only very recently gained much attention. Jurrus et al. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R13" rid="R13" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_320459231" role="button" aria-expanded="false" aria-haspopup="true">13</a>] proposed an automated method to trace axons in serial block-face scanning EM datasets. Their method uses iterative Kalman filtering together with an active contour model and a vector field produced by an optical flow method to estimate the axon location on each slice. Macke et al. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R17" rid="R17" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_232989776" role="button" aria-expanded="false" aria-haspopup="true">17</a>] proposed a probabilistic framework to guide level set propagation on each slice, where the probabilistic framework models the similarity between slices. Mishchenko [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R21" rid="R21" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_232989783" role="button" aria-expanded="false" aria-haspopup="true">21</a>] proposed a 3D neural reconstruction method consisting of a Hessian-based 2D ridge detector to extract axon boundaries and a weighted graph clustering method to generate a connectivity map across slices. All these methods assume that the neural processes follow a specific direction, usually orthogonal to the scanning plane. However, this assumption fails for many axons. The non-axis aligned arbitrary 3D tracking method proposed in this paper provides much more flexibility to handle axons at various orientations. Bartesaghi et al. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R2" rid="R2" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_320459232" role="button" aria-expanded="false" aria-haspopup="true">2</a>] use 3D minimal surface to segment cell boundaries in high resolution electron tomograms. However, their method cannot be directly applied to segment elongated structures as in our case because the method assumes the target structure is spherical topology.</p></div><div id="S4" class="sec"><h3>Vessel Extraction and Virtual Endoscopy</h3><p id="P6">Segmentation and tracking of thin structures is an ongoing area of medical research. Kirbas et al. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R15" rid="R15" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">15</a>] give an extensive review of automated vessel extraction methods. In virtual endoscopy, the focus is on 3D path planning inside elongated structures (e.g., [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R11" rid="R11" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">11</a>, <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R3" rid="R3" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">3</a>, <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R7" rid="R7" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_232989784" role="button" aria-expanded="false" aria-haspopup="true">7</a>]). However, these approaches either require already segmented data or are not directly applicable to high-resolution EM volumes because of the more complicated structures involved.</p></div><div id="S5" class="sec"><h3>Volume Rendering of EM Data</h3><p id="P7">Volume rendering of microscopic structures is a very recent area of research. Mayerich et al. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R20" rid="R20" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_232989779" role="button" aria-expanded="false" aria-haspopup="true">20</a>] segment and visualize microvascular structures and their relationships, but the resolution is two orders of magnitude lower than EM data. We employ GPU-based ray-casting of volumes and implicit surfaces [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R25" rid="R25" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">25</a>] using a bricking scheme for large data [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R4" rid="R4" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">4</a>] implemented in CUDA. Enhancing edges or structure boundaries has always been important in volume rendering, and is typically achieved using higher-order transfer functions [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R14" rid="R14" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">14</a>]. Caban et al. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R5" rid="R5" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_320459233" role="button" aria-expanded="false" aria-haspopup="true">5</a>] have recently introduced texture-based tranfer functions based on first-, second-, and high-order local (histogram) statistics. However, these methods are not effective in dealing with noise in EM images. Our rendering framework employs a general filtering and de-noising step with a neighborhood size that can be changed interactively. Martin et al. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R19" rid="R19" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_320459234" role="button" aria-expanded="false" aria-haspopup="true">19</a>] define a set of brightness, color, and texture cues for constructing a local boundary model. To enhance edges during ray-casting we extended their 2D boundary detection framework using local histogram comparisons.</p></div><div id="S6" class="sec sec-last"><h3>Iso-Surface Rendering</h3><p id="P8">Rendering elongated structures with elliptical cross-sections has been of interest in diffusion tensor imaging (DTI). Interpolation between successive ellipses has been used together with ray-casting [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R24" rid="R24" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">24</a>], as well as with a geometry setup stage [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R23" rid="R23" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_232989775" role="button" aria-expanded="false" aria-haspopup="true">23</a>]. Instead of targeting many relatively thin fibres, we use interpolating quaternion frames between successive ellipses. Our method requires few evaluations of trigonometric functions, which enables high rendering performance and simple implementation.</p></div></div><div id="S7" class="sec"><div class="goto jig-ncbiinpagenav-goto-container"><a class="tgt_dark page-toc-label jig-ncbiinpagenav-goto-heading" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" title="Go to other sections in this page" role="button" aria-expanded="false" aria-haspopup="true">Go to:</a></div><h2 class="head no_bottom_margin ui-helper-clearfix" id="S7title">3 Workflow</h2><p id="P9">We assume that registration of the EM data is performed in a pre-processing step. The individual image tiles acquired by the cameras in the EM are warped and stitched together to obtain slice images of very high resolution. These slices are then registered in 3D so that they are aligned and structures can be followed from slice to slice.</p><div id="S8" class="sec"><h3>3.1 Current Practice</h3><p id="P10">There are a number of manual tools for the segmentations of EM datasets [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R27" rid="R27" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_232989780" role="button" aria-expanded="false" aria-haspopup="true">27</a>]. The software package most commonly used by neuroscientists is <em>Reconstruct</em> [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R9" rid="R9" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_232989777" role="button" aria-expanded="false" aria-haspopup="true">9</a>]. The main window of this tool displays a 2D axis-aligned view of the current slice. To identify the structures of interest, users can move from one slice to the next and inspect each in turn using basic viewing functions such as zoom, pan, scale, and rotation. The segmentation is manual, using polygon, curve, and free-form drawing tools. The final segmented neural processes can be rendered as 3D polygon meshes. To generate higher-quality images, the scientists often use additional volume visualization packages, such as Amira (<a href="http://www.amiravis.com/" ref="reftype=extlink&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CBody&TO=External%7CLink%7CURI&rendering-type=normal" target="pmc_ext">http://www.amiravis.com</a>).</p><p id="P11">This workflow is straightforward but also very labor-intensive and time-consuming. It lacks integrated volume visualization of the input data and high-quality visualization of the resulting segmentation. Reconstruct only allows axis-aligned tracking, so neural processes parallel to the image are difficult to segment. In addition, the data has to fit into main memory, which limits the scalability of the system.</p></div><div id="S9" class="sec sec-last"><h3>3.2 Proposed Workflow</h3><p id="P12"><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F2/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F2" rid-ob="ob-F2" co-legend-rid="lgnd_F2"><span>Figure 2</span></a> illustrates our integrated, interactive workflow for visualizing and segmenting neural processes. The first step in our workflow is to inspect the input volume using volume rendering before any segmentation is performed in order to obtain an overview and to determine a region of interest (ROI) (<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F1/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F1" rid-ob="ob-F1" co-legend-rid="lgnd_F1"><span>Figure 1</span></a> middle). In order to better delineate the structures of interest we modified the volume rendering such that the boundaries of neural processes are depicted more clearly (Section 5).</p><!--fig ft0--><!--fig mode=article f1--><div class="fig iconblock ten_col whole_rhythm clearfix" id="F2" co-legend-rid="lgnd_F2"><a class="icnblk_img figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F2/" target="figure" rid-figpopup="F2" rid-ob="ob-F2"><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/nihms216874f2.gif" class="small-thumb" alt="Fig. 2" title="Fig. 2" src-large="/pmc/articles/PMC3179915/bin/nihms216874f2.jpg"></a><div class="icnblk_cntnt" id="lgnd_F2"><div><a class="figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F2/" target="figure" rid-figpopup="F2" rid-ob="ob-F2">Fig. 2</a></div><!--caption a4--><div><span>Pipeline diagram of our integrated, interactive workflow for visualizing and segmenting neural processes.</span></div></div></div><p id="P13">Using the 3D volume view, the user can specify the center of the current ROI on an arbitrarily oriented 2D clipping plane. The corresponding oblique slice is then shown in an additional 2D view (<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F1/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F1" rid-ob="ob-F1" co-legend-rid="lgnd_F1"><span>Figure 1</span></a> top right). The next step is to quickly paint a rough approximation of a boundary of interest, e.g., of an axon, in this 2D view. This input is used to initialize an active ribbon that automatically performs tracking of the cell boundary from slice to slice (Section 4). The individual cell boundaries are shown in the 2D view and can be inspected and modified interactively at any time.</p><p id="P14">While the segmentation and tracking is in progress, the segmentation obtained thus far is concurrently shown in the 3D volume view (<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F1/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F1" rid-ob="ob-F1" co-legend-rid="lgnd_F1"><span>Figure 1</span></a> middle). The long, elongated structures that are of highest importance in our application can be represented well by elliptical cross-sections. Therefore, in order to use as little memory as possible, an ellipse is fitted to the active ribbon in each slice. These ellipses are interpolated on-the-fly during volume ray-casting in order to obtain smooth, connected implicit surfaces in 3D (Section 5.2). This process is repeated for every neural process of interest, iteratively adding additional structures, which are all depicted concurrently in 3D.</p><p id="P15">The main advantage of our workflow is that it tightly integrates semi-automatic segmentation and visualization, which allows the user to inspect and modify the ongoing tracking and segmentation process at any time, while minimizing the amount of user interaction that is necessary.</p></div></div><div id="S10" class="sec"><div class="goto jig-ncbiinpagenav-goto-container"><a class="tgt_dark page-toc-label jig-ncbiinpagenav-goto-heading" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" title="Go to other sections in this page" role="button" aria-expanded="false" aria-haspopup="true">Go to:</a></div><h2 class="head no_bottom_margin ui-helper-clearfix" id="S10title">4 Segmentation</h2><p id="P16">We compute 3D segmentations using a combination of 2D neural membrane segmentations and 3D centerline tracking. For 2D segmentation we use a modified multiphase level set active ribbon model originally proposed by Vazquez et al. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R32" rid="R32" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">32</a>] (<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F3/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F3" rid-ob="ob-F3" co-legend-rid="lgnd_F3"><span>Figure 3</span></a>). Because level set segmentation is very sensitive to initialization, we propose a novel active ribbon formulation by adding an additional constraint based on image correspondence between current and previous slices. Once 2D segmentation is done, we extrapolate the next point along the centerline of the possibly non-axis aligned neural process.</p><!--fig ft0--><!--fig mode=article f1--><div class="fig iconblock ten_col whole_rhythm clearfix" id="F3" co-legend-rid="lgnd_F3"><a class="icnblk_img figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F3/" target="figure" rid-figpopup="F3" rid-ob="ob-F3"><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/nihms216874f3.gif" class="small-thumb" alt="Fig. 3" title="Fig. 3" src-large="/pmc/articles/PMC3179915/bin/nihms216874f3.jpg"></a><div class="icnblk_cntnt" id="lgnd_F3"><div><a class="figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F3/" target="figure" rid-figpopup="F3" rid-ob="ob-F3">Fig. 3</a></div><!--caption a4--><div><span>Neural process segmentation. Left top: Active ribbon model for 2D neural membrane segmentation. Left bottom: User initialization and solution with inside/outside level sets. Right: 3D centerline tracking.</span></div></div></div><div id="S11" class="sec"><h3>4.1 2D Neural Membrane Segmentation</h3><div id="P17"><p>The <em>active ribbon</em> model is based on two deformable moving interfaces (ϕ<sup>1</sup> and ϕ<sup>2</sup> in <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F3/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F3" rid-ob="ob-F3" co-legend-rid="lgnd_F3"><span>Figure 3</span></a> left) interacting with each other to maintain ribbon topology. The level set equation for each ϕ<sup><em>i</em></sup> of the active ribbon model is defined as follows:
232 </p><div class="disp-formula" id="FD1"><div class="f"><div class="MathJax_Display" role="textbox" aria-readonly="true" style="text-align: center;"><span class="MathJax" id="MathJax-Element-1-Frame" style=""><nobr><span class="math" id="M1" style="width: 20.033em; display: inline-block;"><span style="display: inline-block; position: relative; width: 16.531em; height: 0px; font-size: 121%;"><span style="position: absolute; clip: rect(1.065em 1000.002em 3.583em -0.234em); top: -2.753em; left: 0.002em;"><span class="mrow" id="MathJax-Span-2"><span class="mrow" id="MathJax-Span-3"><span class="mfrac" id="MathJax-Span-4" style="padding-left: 0.12em; padding-right: 0.12em;"><span style="display: inline-block; position: relative; width: 1.576em; height: 0px;"><span style="position: absolute; clip: rect(1.497em 1000.002em 2.835em -0.313em); top: -3.186em; left: 50%; margin-left: -0.706em;"><span class="mrow" id="MathJax-Span-5"><span class="mi" id="MathJax-Span-6" style="font-family: MathJax_Math; font-style: italic;">d<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.002em;"></span></span><span class="msup" id="MathJax-Span-7"><span style="display: inline-block; position: relative; width: 0.907em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.835em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-8" style="font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.556em; left: 0.592em;"><span class="mi" id="MathJax-Span-9" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; clip: rect(1.694em 1000.002em 2.639em -0.352em); top: -1.848em; left: 50%; margin-left: -0.431em;"><span class="mtext" id="MathJax-Span-10" style="font-family: MathJax_Math; font-style: italic;">dt</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; clip: rect(0.868em 1000.002em 1.183em -0.352em); top: -1.297em; left: 0.002em;"><span style="border-left-width: 1.576em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 0.081em; vertical-align: 0.002em;"></span><span style="display: inline-block; width: 0px; height: 1.065em;"></span></span></span></span><span class="mo" id="MathJax-Span-11" style="font-family: MathJax_Main; padding-left: 0.238em;">+</span><span class="mo" id="MathJax-Span-12" style="font-family: MathJax_Main; padding-left: 0.238em;">(</span><span class="mi" id="MathJax-Span-13" style="font-family: MathJax_Math; font-style: italic;">α</span><span class="msub" id="MathJax-Span-14"><span style="display: inline-block; position: relative; width: 1.379em; height: 0px;"><span style="position: absolute; clip: rect(2.009em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.002em;"><span class="mi" id="MathJax-Span-15" style="font-family: MathJax_Main; font-weight: bold;">F</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.71em;"><span class="mi" id="MathJax-Span-16" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">D</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-17" style="font-family: MathJax_Main; padding-left: 0.238em;">+</span><span class="mi" id="MathJax-Span-18" style="font-family: MathJax_Math; font-style: italic; padding-left: 0.238em;">β<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.002em;"></span></span><span class="msub" id="MathJax-Span-19"><span style="display: inline-block; position: relative; width: 1.34em; height: 0px;"><span style="position: absolute; clip: rect(2.009em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.002em;"><span class="mi" id="MathJax-Span-20" style="font-family: MathJax_Main; font-weight: bold;">F</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.71em;"><span class="mi" id="MathJax-Span-21" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">R</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-22" style="font-family: MathJax_Main; padding-left: 0.238em;">+</span><span class="mi" id="MathJax-Span-23" style="font-family: MathJax_Math; font-style: italic; padding-left: 0.238em;">γ<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.041em;"></span></span><span class="msub" id="MathJax-Span-24"><span style="display: inline-block; position: relative; width: 1.419em; height: 0px;"><span style="position: absolute; clip: rect(2.009em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.002em;"><span class="mi" id="MathJax-Span-25" style="font-family: MathJax_Main; font-weight: bold;">F</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.71em;"><span class="mi" id="MathJax-Span-26" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">K<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.041em;"></span></span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-27" style="font-family: MathJax_Main;">)</span><span class="mo" id="MathJax-Span-28" style="font-family: MathJax_Main;">|</span><span class="mo" id="MathJax-Span-29" style="font-family: MathJax_Main;">∇</span><span class="msup" id="MathJax-Span-30"><span style="display: inline-block; position: relative; width: 0.907em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.835em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-31" style="font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.556em; left: 0.592em;"><span class="mi" id="MathJax-Span-32" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-33" style="font-family: MathJax_Main;">|</span><span class="mo" id="MathJax-Span-34" style="font-family: MathJax_Main; padding-left: 0.277em;">=</span><span class="mn" id="MathJax-Span-35" style="font-family: MathJax_Main; padding-left: 0.277em;">0</span><span class="mo" id="MathJax-Span-36" style="font-family: MathJax_Main;">,</span></span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span><span style="border-left-width: 0.002em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 2.812em; vertical-align: -0.902em;"></span></span></nobr></span></div><script type="math/mml" id="MathJax-Element-1"><math id="M1" display="block" overflow="scroll"><mrow><mfrac><mrow><mi>d</mi><msup><mi>ϕ</mi><mi>i</mi></msup></mrow><mtext mathvariant="italic">dt</mtext></mfrac><mo>+</mo><mo stretchy="false">(</mo><mi>α</mi><msub><mi mathvariant="bold">F</mi><mi>D</mi></msub><mo>+</mo><mi>β</mi><msub><mi mathvariant="bold">F</mi><mi>R</mi></msub><mo>+</mo><mi>γ</mi><msub><mi mathvariant="bold">F</mi><mi>K</mi></msub><mo stretchy="false">)</mo><mo stretchy="false">|</mo><mo>∇</mo><msup><mi>ϕ</mi><mi>i</mi></msup><mo stretchy="false">|</mo><mo>=</mo><mn>0</mn><mo>,</mo></mrow></math></script></div><div class="l">(1)</div></div><p>
234 </p><div class="disp-formula" id="FD2"><div class="f"><div class="MathJax_Display" role="textbox" aria-readonly="true" style="text-align: center;"><span class="MathJax" id="MathJax-Element-2-Frame" style=""><nobr><span class="math" id="M2" style="width: 16.806em; display: inline-block;"><span style="display: inline-block; position: relative; width: 13.894em; height: 0px; font-size: 121%;"><span style="position: absolute; clip: rect(2.363em 1000.002em 5em -0.313em); top: -3.933em; left: 0.002em;"><span class="mrow" id="MathJax-Span-38"><span class="mrow" id="MathJax-Span-39"><span class="msub" id="MathJax-Span-40"><span style="display: inline-block; position: relative; width: 1.379em; height: 0px;"><span style="position: absolute; clip: rect(2.009em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.002em;"><span class="mi" id="MathJax-Span-41" style="font-family: MathJax_Main; font-weight: bold;">F</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.71em;"><span class="mi" id="MathJax-Span-42" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">D</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-43" style="font-family: MathJax_Main; padding-left: 0.277em;">=</span><span class="mo" id="MathJax-Span-44" style="font-family: MathJax_Main; padding-left: 0.277em;">(</span><span class="msub" id="MathJax-Span-45"><span style="display: inline-block; position: relative; width: 0.868em; height: 0px;"><span style="position: absolute; clip: rect(1.97em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-46" style="font-family: MathJax_Math; font-style: italic;">c</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.123em; left: 0.435em;"><span class="mn" id="MathJax-Span-47" style="font-size: 70.7%; font-family: MathJax_Main;">2</span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-48" style="font-family: MathJax_Main; padding-left: 0.238em;">−</span><span class="msub" id="MathJax-Span-49" style="padding-left: 0.238em;"><span style="display: inline-block; position: relative; width: 0.868em; height: 0px;"><span style="position: absolute; clip: rect(1.97em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-50" style="font-family: MathJax_Math; font-style: italic;">c</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.123em; left: 0.435em;"><span class="mn" id="MathJax-Span-51" style="font-size: 70.7%; font-family: MathJax_Main;">1</span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-52" style="font-family: MathJax_Main;">)</span><span class="mspace" id="MathJax-Span-53" style="height: 0.002em; vertical-align: 0.002em; width: 0.159em; display: inline-block; overflow: hidden;"></span><span class="mo" id="MathJax-Span-54" style="vertical-align: 0.002em;"><span style="font-family: MathJax_Size3;">(</span></span><span class="mi" id="MathJax-Span-55" style="font-family: MathJax_Math; font-style: italic;">I<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.081em;"></span></span><span class="mo" id="MathJax-Span-56" style="font-family: MathJax_Main; padding-left: 0.238em;">−</span><span class="mfrac" id="MathJax-Span-57" style="padding-left: 0.356em; padding-right: 0.12em;"><span style="display: inline-block; position: relative; width: 3.898em; height: 0px;"><span style="position: absolute; clip: rect(1.891em 1000.002em 3.111em -0.274em); top: -3.422em; left: 50%; margin-left: -1.887em;"><span class="mrow" id="MathJax-Span-58"><span class="mo" id="MathJax-Span-59" style="font-family: MathJax_Main;">(</span><span class="msub" id="MathJax-Span-60"><span style="display: inline-block; position: relative; width: 0.868em; height: 0px;"><span style="position: absolute; clip: rect(1.97em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-61" style="font-family: MathJax_Math; font-style: italic;">c</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.123em; left: 0.435em;"><span class="mn" id="MathJax-Span-62" style="font-size: 70.7%; font-family: MathJax_Main;">1</span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-63" style="font-family: MathJax_Main; padding-left: 0.238em;">+</span><span class="msub" id="MathJax-Span-64" style="padding-left: 0.238em;"><span style="display: inline-block; position: relative; width: 0.868em; height: 0px;"><span style="position: absolute; clip: rect(1.97em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-65" style="font-family: MathJax_Math; font-style: italic;">c</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.123em; left: 0.435em;"><span class="mn" id="MathJax-Span-66" style="font-size: 70.7%; font-family: MathJax_Main;">2</span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-67" style="font-family: MathJax_Main;">)</span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span><span style="position: absolute; clip: rect(1.97em 1000.002em 2.875em -0.313em); top: -2.084em; left: 50%; margin-left: -0.274em;"><span class="mn" id="MathJax-Span-68" style="font-family: MathJax_Main;">2</span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span><span style="position: absolute; clip: rect(0.868em 1000.002em 1.183em -0.352em); top: -1.297em; left: 0.002em;"><span style="border-left-width: 3.898em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 0.081em; vertical-align: 0.002em;"></span><span style="display: inline-block; width: 0px; height: 1.065em;"></span></span></span></span><span class="mo" id="MathJax-Span-69" style="vertical-align: 0.002em;"><span style="font-family: MathJax_Size3;">)</span></span></span></span><span style="display: inline-block; width: 0px; height: 3.937em;"></span></span></span><span style="border-left-width: 0.002em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 3.002em; vertical-align: -1.188em;"></span></span></nobr></span></div><script type="math/mml" id="MathJax-Element-2"><math id="M2" display="block" overflow="scroll"><mrow><msub><mi mathvariant="bold">F</mi><mi>D</mi></msub><mo>=</mo><mo stretchy="false">(</mo><msub><mi>c</mi><mn>2</mn></msub><mo>−</mo><msub><mi>c</mi><mn>1</mn></msub><mo stretchy="false">)</mo><mspace width="thinmathspace"></mspace><mo stretchy="true">(</mo><mi>I</mi><mo>−</mo><mfrac><mrow><mo stretchy="false">(</mo><msub><mi>c</mi><mn>1</mn></msub><mo>+</mo><msub><mi>c</mi><mn>2</mn></msub><mo stretchy="false">)</mo></mrow><mn>2</mn></mfrac><mo stretchy="true">)</mo></mrow></math></script></div></div><p>
235 is the data dependent speed to move toward the membrane boundary, <em>c</em><sub>1</sub> and <em>c</em><sub>2</sub> are the average pixel intensity of inner cell region and cell membrane, respectively, and <em>I</em> is the pixel intensity of the input image.
236 </p><div class="disp-formula" id="FD3"><div class="f"><div class="MathJax_Display" role="textbox" aria-readonly="true" style="text-align: center;"><span class="MathJax" id="MathJax-Element-3-Frame" style=""><nobr><span class="math" id="M3" style="width: 12.871em; display: inline-block;"><span style="display: inline-block; position: relative; width: 10.628em; height: 0px; font-size: 121%;"><span style="position: absolute; clip: rect(1.065em 1000.002em 3.898em -0.313em); top: -2.753em; left: 0.002em;"><span class="mrow" id="MathJax-Span-71"><span class="mrow" id="MathJax-Span-72"><span class="msub" id="MathJax-Span-73"><span style="display: inline-block; position: relative; width: 1.34em; height: 0px;"><span style="position: absolute; clip: rect(2.009em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.002em;"><span class="mi" id="MathJax-Span-74" style="font-family: MathJax_Main; font-weight: bold;">F</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.71em;"><span class="mi" id="MathJax-Span-75" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">R</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-76" style="font-family: MathJax_Main; padding-left: 0.277em;">=</span><span class="msub" id="MathJax-Span-77" style="padding-left: 0.277em;"><span style="display: inline-block; position: relative; width: 0.907em; height: 0px;"><span style="position: absolute; clip: rect(1.97em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-78" style="font-family: MathJax_Math; font-style: italic;">σ<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.002em;"></span></span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.592em;"><span class="mi" id="MathJax-Span-79" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-80" style="font-family: MathJax_Main;">(</span><span class="msup" id="MathJax-Span-81"><span style="display: inline-block; position: relative; width: 0.946em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.835em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-82" style="font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.556em; left: 0.592em;"><span class="mi" id="MathJax-Span-83" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">j</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-84" style="font-family: MathJax_Main;">)</span><span class="mo" id="MathJax-Span-85" style="font-family: MathJax_Main;">∇</span><span class="msup" id="MathJax-Span-86"><span style="display: inline-block; position: relative; width: 0.946em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.835em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-87" style="font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.556em; left: 0.592em;"><span class="mi" id="MathJax-Span-88" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">j</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-89" style="font-family: MathJax_Main; padding-left: 0.238em;">⋅</span><span class="mfrac" id="MathJax-Span-90" style="padding-left: 0.356em; padding-right: 0.12em;"><span style="display: inline-block; position: relative; width: 2.56em; height: 0px;"><span style="position: absolute; clip: rect(1.734em 1000.002em 3.072em -0.313em); top: -3.422em; left: 50%; margin-left: -0.864em;"><span class="mrow" id="MathJax-Span-91"><span class="mo" id="MathJax-Span-92" style="font-family: MathJax_Main;">∇</span><span class="msup" id="MathJax-Span-93"><span style="display: inline-block; position: relative; width: 0.907em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.835em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-94" style="font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.556em; left: 0.592em;"><span class="mi" id="MathJax-Span-95" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span><span style="position: absolute; clip: rect(1.773em 1000.002em 3.111em -0.077em); top: -1.966em; left: 50%; margin-left: -1.218em;"><span class="mrow" id="MathJax-Span-96"><span class="mo" id="MathJax-Span-97" style="font-family: MathJax_Size1;">‖</span><span class="mrow" id="MathJax-Span-98"><span class="msup" id="MathJax-Span-99"><span style="display: inline-block; position: relative; width: 0.907em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.835em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-100" style="font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.556em; left: 0.592em;"><span class="mi" id="MathJax-Span-101" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span></span><span class="mo" id="MathJax-Span-102" style="font-family: MathJax_Size1;">‖</span></span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; clip: rect(0.868em 1000.002em 1.183em -0.352em); top: -1.297em; left: 0.002em;"><span style="border-left-width: 2.56em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 0.081em; vertical-align: 0.002em;"></span><span style="display: inline-block; width: 0px; height: 1.065em;"></span></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span><span style="border-left-width: 0.002em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 3.24em; vertical-align: -1.283em;"></span></span></nobr></span></div><script type="math/mml" id="MathJax-Element-3"><math id="M3" display="block" overflow="scroll"><mrow><msub><mi mathvariant="bold">F</mi><mi>R</mi></msub><mo>=</mo><msub><mi>σ</mi><mi>i</mi></msub><mo stretchy="false">(</mo><msup><mi>ϕ</mi><mi>j</mi></msup><mo stretchy="false">)</mo><mo>∇</mo><msup><mi>ϕ</mi><mi>j</mi></msup><mo>·</mo><mfrac><mrow><mo>∇</mo><msup><mi>ϕ</mi><mi>i</mi></msup></mrow><mrow><mo stretchy="false">‖</mo><mrow><msup><mi>ϕ</mi><mi>i</mi></msup></mrow><mo stretchy="false">‖</mo></mrow></mfrac></mrow></math></script></div></div><p>
237 is the ribbon consistency speed to keep constant distance between two level set interfaces ϕ<sup><em>i</em></sup> and ϕ<sup><em>j</em></sup>. σ<sub><em>i</em></sub> (ϕ<sup><em>j</em></sup>) returns a positive value if two level sets are too close and a negative value otherwise (more details can be found in [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R32" rid="R32" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">32</a>]). <strong>F</strong><sub><em>K</em></sub> is the mean curvature speed to maintain the smoothness of the interfaces. Because two interfaces push or pull each other until they converge to the target, the active ribbon model is very robust in noisy and feature-rich EM images. In addition, the model includes a force field that allows neighboring ribbons to interact with each other.</p></div><p id="P18">However, Vazques et al. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R32" rid="R32" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">32</a>] assume that neural processes are orthogonal to the image plane and that there is no large displacement between consecutive slices, which is typically not the case. In addition, the cross sectional shape of the neural process may deform significantly between slices, making initialization of the active ribbon challenging. Therefore, we propose a new active ribbon formulation by adding a force field that maps one image to another using image correspondence. This allows us to robustly initialize the location of the neural membranes on subsequent slices.</p><div id="P19"><p>Let <em>I<sub>i</sub></em> and <em>I</em><sub><em>i</em>+1</sub> be two consecutive slices, where each slice is defined on a 2D domain Ω. We can define the energy between two images for a given vector field <strong>v</strong>, which describes how <em>different</em> two images are, as follows:
238 </p><div class="disp-formula" id="FD4"><div class="f"><div class="MathJax_Display" role="textbox" aria-readonly="true" style="text-align: center;"><span class="MathJax" id="MathJax-Element-4-Frame" style=""><nobr><span class="math" id="M4" style="width: 19.915em; display: inline-block;"><span style="display: inline-block; position: relative; width: 16.452em; height: 0px; font-size: 121%;"><span style="position: absolute; clip: rect(1.261em 1000.002em 3.78em -0.313em); top: -2.753em; left: 0.002em;"><span class="mrow" id="MathJax-Span-104"><span class="mrow" id="MathJax-Span-105"><span class="msub" id="MathJax-Span-106"><span style="display: inline-block; position: relative; width: 1.183em; height: 0px;"><span style="position: absolute; clip: rect(1.734em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-107" style="font-family: MathJax_Math; font-style: italic;">E<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.041em;"></span></span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.75em;"><span class="mi" id="MathJax-Span-108" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">I<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.041em;"></span></span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-109" style="font-family: MathJax_Main; padding-left: 0.277em;">=</span><span class="mfrac" id="MathJax-Span-110" style="padding-left: 0.396em; padding-right: 0.12em;"><span style="display: inline-block; position: relative; width: 0.632em; height: 0px;"><span style="position: absolute; clip: rect(1.97em 1000.002em 2.875em -0.274em); top: -3.422em; left: 50%; margin-left: -0.274em;"><span class="mn" id="MathJax-Span-111" style="font-family: MathJax_Main;">1</span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span><span style="position: absolute; clip: rect(1.97em 1000.002em 2.875em -0.313em); top: -2.084em; left: 50%; margin-left: -0.274em;"><span class="mn" id="MathJax-Span-112" style="font-family: MathJax_Main;">2</span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span><span style="position: absolute; clip: rect(0.868em 1000.002em 1.183em -0.352em); top: -1.297em; left: 0.002em;"><span style="border-left-width: 0.632em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 0.081em; vertical-align: 0.002em;"></span><span style="display: inline-block; width: 0px; height: 1.065em;"></span></span></span></span><span class="mstyle" id="MathJax-Span-113" style="padding-left: 0.159em;"><span class="mrow" id="MathJax-Span-114"><span class="mrow" id="MathJax-Span-115"><span class="msub" id="MathJax-Span-116"><span style="display: inline-block; position: relative; width: 2.088em; height: 0px;"><span style="position: absolute; clip: rect(2.245em 1000.002em 4.725em -0.313em); top: -3.737em; left: 0.002em;"><span class="mo" id="MathJax-Span-117" style="font-family: MathJax_Size2; vertical-align: 0.002em;">∫<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.396em;"></span></span><span style="display: inline-block; width: 0px; height: 3.741em;"></span></span><span style="position: absolute; top: -1.415em; left: 0.592em;"><span class="mrow" id="MathJax-Span-118"><span class="mi" id="MathJax-Span-119" style="font-size: 70.7%; font-family: MathJax_Main; font-weight: bold;">x</span><span class="mo" id="MathJax-Span-120" style="font-size: 70.7%; font-family: MathJax_Main;">∈</span><span class="mi" id="MathJax-Span-121" style="font-size: 70.7%; font-family: MathJax_Main;">Ω</span></span><span style="display: inline-block; width: 0px; height: 2.324em;"></span></span></span></span></span></span></span><span class="msup" id="MathJax-Span-122" style="padding-left: 0.159em;"><span style="display: inline-block; position: relative; width: 5.197em; height: 0px;"><span style="position: absolute; clip: rect(1.773em 1000.002em 3.111em -0.274em); top: -2.753em; left: 0.002em;"><span class="mrow" id="MathJax-Span-123"><span class="mo" id="MathJax-Span-124" style="font-family: MathJax_Main;">(</span><span class="msub" id="MathJax-Span-125"><span style="display: inline-block; position: relative; width: 1.025em; height: 0px;"><span style="position: absolute; clip: rect(3.032em 1000.002em 4.134em -0.313em); top: -4.012em; left: 0.002em;"><span class="mover" id="MathJax-Span-126"><span style="display: inline-block; position: relative; width: 0.71em; height: 0px;"><span style="position: absolute; clip: rect(1.734em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-127" style="font-family: MathJax_Math; font-style: italic;">I<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.081em;"></span></span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; clip: rect(1.97em 1000.002em 2.324em -0.274em); top: -2.95em; left: 0.199em;"><span class="mo" id="MathJax-Span-128" style=""><span style="font-family: MathJax_Main;">˜</span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span></span><span style="display: inline-block; width: 0px; height: 4.016em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.71em;"><span class="mi" id="MathJax-Span-129" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-130" style="font-family: MathJax_Main; padding-left: 0.238em;">−</span><span class="msub" id="MathJax-Span-131" style="padding-left: 0.238em;"><span style="display: inline-block; position: relative; width: 1.694em; height: 0px;"><span style="position: absolute; clip: rect(1.734em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-132" style="font-family: MathJax_Math; font-style: italic;">I<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.081em;"></span></span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.123em; left: 0.474em;"><span class="mrow" id="MathJax-Span-133"><span class="mi" id="MathJax-Span-134" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span class="mo" id="MathJax-Span-135" style="font-size: 70.7%; font-family: MathJax_Main;">+</span><span class="mn" id="MathJax-Span-136" style="font-size: 70.7%; font-family: MathJax_Main;">1</span></span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-137" style="font-family: MathJax_Main;">)</span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span><span style="position: absolute; top: -2.91em; left: 4.764em;"><span class="mn" id="MathJax-Span-138" style="font-size: 70.7%; font-family: MathJax_Main;">2</span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-139" style="font-family: MathJax_Main; padding-left: 0.238em;">+</span><span class="mi" id="MathJax-Span-140" style="font-family: MathJax_Math; font-style: italic; padding-left: 0.238em;">α</span><span class="msup" id="MathJax-Span-141"><span style="display: inline-block; position: relative; width: 3.347em; height: 0px;"><span style="position: absolute; clip: rect(2.088em 1000.002em 3.032em -0.077em); top: -2.871em; left: 0.002em;"><span class="mrow" id="MathJax-Span-142"><span class="mo" id="MathJax-Span-143" style="font-family: MathJax_Size1;">‖</span><span class="mrow" id="MathJax-Span-144"><span class="mo" id="MathJax-Span-145" style="font-family: MathJax_Main;">∇</span><span class="mi" id="MathJax-Span-146" style="font-family: MathJax_Main; font-weight: bold;">v</span></span><span class="mo" id="MathJax-Span-147" style="font-family: MathJax_Size1;">‖</span></span><span style="display: inline-block; width: 0px; height: 2.875em;"></span></span><span style="position: absolute; top: -2.674em; left: 2.914em;"><span class="mn" id="MathJax-Span-148" style="font-size: 70.7%; font-family: MathJax_Main;">2</span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-149" style="font-family: MathJax_Main;">,</span></span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span><span style="border-left-width: 0.002em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 2.86em; vertical-align: -1.14em;"></span></span></nobr></span></div><script type="math/mml" id="MathJax-Element-4"><math id="M4" display="block" overflow="scroll"><mrow><msub><mi>E</mi><mi>I</mi></msub><mo>=</mo><mfrac><mn>1</mn><mn>2</mn></mfrac><mstyle displaystyle="true"><mrow><msub><mo stretchy="true">∫</mo><mrow><mi mathvariant="bold">x</mi><mo>∈</mo><mi mathvariant="normal">Ω</mi></mrow></msub></mrow></mstyle><msup><mrow><mo stretchy="false">(</mo><msub><mover accent="true"><mi>I</mi><mo>~</mo></mover><mi>i</mi></msub><mo>−</mo><msub><mi>I</mi><mrow><mi>i</mi><mo>+</mo><mn>1</mn></mrow></msub><mo stretchy="false">)</mo></mrow><mn>2</mn></msup><mo>+</mo><mi>α</mi><msup><mrow><mo stretchy="false">‖</mo><mrow><mo>∇</mo><mi mathvariant="bold">v</mi></mrow><mo stretchy="false">‖</mo></mrow><mn>2</mn></msup><mo>,</mo></mrow></math></script></div><div class="l">(2)</div></div><p>
239 where <em>Ĩ<sub>i</sub></em> is the image <em>I<sub>i</sub></em> deformed by the vector field <strong>v</strong>, and α is a regularization parameter. Finding the vector field <strong>v</strong> that minimizes <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#FD4" rid="FD4" class=" disp-formula">Equation 2</a> is a nonrigid image registration problem [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R1" rid="R1" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">1</a>, <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R6" rid="R6" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">6</a>]. In order to find <strong>v</strong>, we use a gradient flow method along the negative gradient direction of <em>E<sub>I</sub></em> with respect to <strong>v</strong>. To avoid local minima and to handle large deformations more efficiently, we use a multilevel approach and compute the solution on different scales, from coarse to fine.</p></div><div id="P20"><p>Once we have the vector field <strong>v</strong>, we can define the energy <em>E</em><sub>ϕ</sub> that measures the difference between two distance fields ϕ<sub><em>i</em></sub> and ϕ<sub><em>i</em>+1</sub> for the images <em>I<sub>i</sub></em> and <em>I</em><sub><em>i</em>+1</sub> as follows:
240 </p><div class="disp-formula" id="FD5"><div class="f"><div class="MathJax_Display" role="textbox" aria-readonly="true" style="text-align: center;"><span class="MathJax" id="MathJax-Element-5-Frame" style=""><nobr><span class="math" id="M5" style="width: 12.202em; display: inline-block;"><span style="display: inline-block; position: relative; width: 10.077em; height: 0px; font-size: 121%;"><span style="position: absolute; clip: rect(1.261em 1000.002em 3.78em -0.313em); top: -2.753em; left: 0.002em;"><span class="mrow" id="MathJax-Span-151"><span class="mrow" id="MathJax-Span-152"><span class="msub" id="MathJax-Span-153"><span style="display: inline-block; position: relative; width: 1.261em; height: 0px;"><span style="position: absolute; clip: rect(1.734em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-154" style="font-family: MathJax_Math; font-style: italic;">E<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.041em;"></span></span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.75em;"><span class="mi" id="MathJax-Span-155" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-156" style="font-family: MathJax_Main; padding-left: 0.277em;">=</span><span class="mstyle" id="MathJax-Span-157" style="padding-left: 0.277em;"><span class="mrow" id="MathJax-Span-158"><span class="mrow" id="MathJax-Span-159"><span class="msub" id="MathJax-Span-160"><span style="display: inline-block; position: relative; width: 2.088em; height: 0px;"><span style="position: absolute; clip: rect(2.245em 1000.002em 4.725em -0.313em); top: -3.737em; left: 0.002em;"><span class="mo" id="MathJax-Span-161" style="font-family: MathJax_Size2; vertical-align: 0.002em;">∫<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.396em;"></span></span><span style="display: inline-block; width: 0px; height: 3.741em;"></span></span><span style="position: absolute; top: -1.415em; left: 0.592em;"><span class="mrow" id="MathJax-Span-162"><span class="mi" id="MathJax-Span-163" style="font-size: 70.7%; font-family: MathJax_Main; font-weight: bold;">x</span><span class="mo" id="MathJax-Span-164" style="font-size: 70.7%; font-family: MathJax_Main;">∈</span><span class="mi" id="MathJax-Span-165" style="font-size: 70.7%; font-family: MathJax_Main;">Ω</span></span><span style="display: inline-block; width: 0px; height: 2.324em;"></span></span></span></span></span></span></span><span class="mo" id="MathJax-Span-166" style="font-family: MathJax_Main; padding-left: 0.159em;">|</span><span class="msub" id="MathJax-Span-167"><span style="display: inline-block; position: relative; width: 1.812em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.835em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-168" style="font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.044em; left: 0.592em;"><span class="mrow" id="MathJax-Span-169"><span class="mi" id="MathJax-Span-170" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span class="mo" id="MathJax-Span-171" style="font-size: 70.7%; font-family: MathJax_Main;">+</span><span class="mn" id="MathJax-Span-172" style="font-size: 70.7%; font-family: MathJax_Main;">1</span></span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-173" style="font-family: MathJax_Main; padding-left: 0.238em;">−</span><span class="msub" id="MathJax-Span-174" style="padding-left: 0.238em;"><span style="display: inline-block; position: relative; width: 0.946em; height: 0px;"><span style="position: absolute; clip: rect(2.993em 1000.002em 4.331em -0.313em); top: -4.012em; left: 0.002em;"><span class="mover" id="MathJax-Span-175"><span style="display: inline-block; position: relative; width: 0.632em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.835em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-176" style="font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; clip: rect(1.97em 1000.002em 2.324em -0.274em); top: -2.989em; left: 0.12em;"><span class="mo" id="MathJax-Span-177" style=""><span style="font-family: MathJax_Main;">˜</span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span></span><span style="display: inline-block; width: 0px; height: 4.016em;"></span></span><span style="position: absolute; top: -1.887em; left: 0.632em;"><span class="mi" id="MathJax-Span-178" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="msup" id="MathJax-Span-179"><span style="display: inline-block; position: relative; width: 0.71em; height: 0px;"><span style="position: absolute; clip: rect(1.891em 1000.002em 3.111em -0.234em); top: -2.753em; left: 0.002em;"><span class="mo" id="MathJax-Span-180" style="font-family: MathJax_Main;">|</span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span><span style="position: absolute; top: -2.753em; left: 0.277em;"><span class="mn" id="MathJax-Span-181" style="font-size: 70.7%; font-family: MathJax_Main;">2</span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-182" style="font-family: MathJax_Main;">,</span></span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span><span style="border-left-width: 0.002em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 2.86em; vertical-align: -1.14em;"></span></span></nobr></span></div><script type="math/mml" id="MathJax-Element-5"><math id="M5" display="block" overflow="scroll"><mrow><msub><mi>E</mi><mi>ϕ</mi></msub><mo>=</mo><mstyle displaystyle="true"><mrow><msub><mo stretchy="true">∫</mo><mrow><mi mathvariant="bold">x</mi><mo>∈</mo><mi mathvariant="normal">Ω</mi></mrow></msub></mrow></mstyle><mo stretchy="false">|</mo><msub><mi>ϕ</mi><mrow><mi>i</mi><mo>+</mo><mn>1</mn></mrow></msub><mo>−</mo><msub><mover accent="true"><mi>ϕ</mi><mo>~</mo></mover><mi>i</mi></msub><msup><mo stretchy="false">|</mo><mn>2</mn></msup><mo>,</mo></mrow></math></script></div><div class="l">(3)</div></div><p>
241 where <img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/x03D5x0303.gif" border="0" alt="[phi with tilde]" title=""><sub><em>i</em></sub> is ϕ<sub><em>i</em></sub> deformed by <strong>v</strong>. Note that ϕ<sub><em>i</em></sub> and ϕ<sub><em>i</em>+1</sub> in this discussion are not the inner and outer level set for the active ribbon (<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F3/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F3" rid-ob="ob-F3" co-legend-rid="lgnd_F3"><span>Figure 3</span></a> left) but a single level set on two different images. Thus we can define the level set function for ϕ<sub><em>i</em>+1</sub> that minimizes <em>E</em><sub>ϕ</sub> as follows:
242 </p><div class="disp-formula" id="FD6"><div class="f"><div class="MathJax_Display" role="textbox" aria-readonly="true" style="text-align: center;"><span class="MathJax" id="MathJax-Element-6-Frame" style=""><nobr><span class="math" id="M6" style="width: 13.54em; display: inline-block;"><span style="display: inline-block; position: relative; width: 11.179em; height: 0px; font-size: 121%;"><span style="position: absolute; clip: rect(1.222em 1000.002em 3.583em -0.234em); top: -2.753em; left: 0.002em;"><span class="mrow" id="MathJax-Span-184"><span class="mrow" id="MathJax-Span-185"><span class="mfrac" id="MathJax-Span-186" style="padding-left: 0.12em; padding-right: 0.12em;"><span style="display: inline-block; position: relative; width: 2.481em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.954em -0.313em); top: -3.264em; left: 50%; margin-left: -1.179em;"><span class="mrow" id="MathJax-Span-187"><span class="mi" id="MathJax-Span-188" style="font-family: MathJax_Math; font-style: italic;">d<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.002em;"></span></span><span class="msub" id="MathJax-Span-189"><span style="display: inline-block; position: relative; width: 1.812em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.835em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-190" style="font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.044em; left: 0.592em;"><span class="mrow" id="MathJax-Span-191"><span class="mi" id="MathJax-Span-192" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span class="mo" id="MathJax-Span-193" style="font-size: 70.7%; font-family: MathJax_Main;">+</span><span class="mn" id="MathJax-Span-194" style="font-size: 70.7%; font-family: MathJax_Main;">1</span></span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; clip: rect(1.694em 1000.002em 2.639em -0.352em); top: -1.848em; left: 50%; margin-left: -0.431em;"><span class="mtext" id="MathJax-Span-195" style="font-family: MathJax_Math; font-style: italic;">dt</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; clip: rect(0.868em 1000.002em 1.183em -0.352em); top: -1.297em; left: 0.002em;"><span style="border-left-width: 2.481em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 0.081em; vertical-align: 0.002em;"></span><span style="display: inline-block; width: 0px; height: 1.065em;"></span></span></span></span><span class="mo" id="MathJax-Span-196" style="font-family: MathJax_Main; padding-left: 0.238em;">+</span><span class="mi" id="MathJax-Span-197" style="font-family: MathJax_Math; font-style: italic; padding-left: 0.238em;">ξ<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.002em;"></span></span><span class="mspace" id="MathJax-Span-198" style="height: 0.002em; vertical-align: 0.002em; width: 0.159em; display: inline-block; overflow: hidden;"></span><span class="msub" id="MathJax-Span-199"><span style="display: inline-block; position: relative; width: 1.34em; height: 0px;"><span style="position: absolute; clip: rect(2.009em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.002em;"><span class="mi" id="MathJax-Span-200" style="font-family: MathJax_Main; font-weight: bold;">F</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.71em;"><span class="mi" id="MathJax-Span-201" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">C<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.041em;"></span></span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-202" style="font-family: MathJax_Main;">|</span><span class="mo" id="MathJax-Span-203" style="font-family: MathJax_Main;">∇</span><span class="msub" id="MathJax-Span-204"><span style="display: inline-block; position: relative; width: 1.812em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.835em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-205" style="font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.044em; left: 0.592em;"><span class="mrow" id="MathJax-Span-206"><span class="mi" id="MathJax-Span-207" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span class="mo" id="MathJax-Span-208" style="font-size: 70.7%; font-family: MathJax_Main;">+</span><span class="mn" id="MathJax-Span-209" style="font-size: 70.7%; font-family: MathJax_Main;">1</span></span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-210" style="font-family: MathJax_Main;">|</span><span class="mo" id="MathJax-Span-211" style="font-family: MathJax_Main; padding-left: 0.277em;">=</span><span class="mn" id="MathJax-Span-212" style="font-family: MathJax_Main; padding-left: 0.277em;">0</span><span class="mo" id="MathJax-Span-213" style="font-family: MathJax_Main;">,</span></span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span><span style="border-left-width: 0.002em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 2.669em; vertical-align: -0.902em;"></span></span></nobr></span></div><script type="math/mml" id="MathJax-Element-6"><math id="M6" display="block" overflow="scroll"><mrow><mfrac><mrow><mi>d</mi><msub><mi>ϕ</mi><mrow><mi>i</mi><mo>+</mo><mn>1</mn></mrow></msub></mrow><mtext mathvariant="italic">dt</mtext></mfrac><mo>+</mo><mi>ξ</mi><mspace width="thinmathspace"></mspace><msub><mi mathvariant="bold">F</mi><mi>C</mi></msub><mo stretchy="false">|</mo><mo>∇</mo><msub><mi>ϕ</mi><mrow><mi>i</mi><mo>+</mo><mn>1</mn></mrow></msub><mo stretchy="false">|</mo><mo>=</mo><mn>0</mn><mo>,</mo></mrow></math></script></div><div class="l">(4)</div></div><p>
243 where <strong>F</strong><sub><em>C</em></sub> is the image correspondence speed and γ is a level set parameter. The image correspondence speed <strong>F</strong><sub><em>C</em></sub> can be defined using the gradient of <em>E</em><sub>ϕ</sub> with respect to ϕ as follows:
244 </p><div class="disp-formula" id="FD7"><div class="f"><div class="MathJax_Display" role="textbox" aria-readonly="true" style="text-align: center;"><span class="MathJax" id="MathJax-Element-7-Frame" style=""><nobr><span class="math" id="M7" style="width: 17.593em; display: inline-block;"><span style="display: inline-block; position: relative; width: 14.524em; height: 0px; font-size: 121%;"><span style="position: absolute; clip: rect(1.025em 1000.002em 3.859em -0.313em); top: -2.753em; left: 0.002em;"><span class="mrow" id="MathJax-Span-215"><span class="mrow" id="MathJax-Span-216"><span class="msub" id="MathJax-Span-217"><span style="display: inline-block; position: relative; width: 1.34em; height: 0px;"><span style="position: absolute; clip: rect(2.009em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.002em;"><span class="mi" id="MathJax-Span-218" style="font-family: MathJax_Main; font-weight: bold;">F</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.71em;"><span class="mi" id="MathJax-Span-219" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">C<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.041em;"></span></span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-220" style="font-family: MathJax_Main; padding-left: 0.277em;">=</span><span class="mfrac" id="MathJax-Span-221" style="padding-left: 0.396em; padding-right: 0.12em;"><span style="display: inline-block; position: relative; width: 11.179em; height: 0px;"><span style="position: absolute; clip: rect(1.734em 1000.002em 3.19em -0.352em); top: -3.501em; left: 50%; margin-left: -5.508em;"><span class="mrow" id="MathJax-Span-222"><span class="mtext" id="MathJax-Span-223" style="font-family: MathJax_Main;">sign</span><span class="mo" id="MathJax-Span-224" style="font-family: MathJax_Main;">(</span><span class="msub" id="MathJax-Span-225"><span style="display: inline-block; position: relative; width: 1.812em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.835em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-226" style="font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.044em; left: 0.592em;"><span class="mrow" id="MathJax-Span-227"><span class="mi" id="MathJax-Span-228" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span class="mo" id="MathJax-Span-229" style="font-size: 70.7%; font-family: MathJax_Main;">+</span><span class="mn" id="MathJax-Span-230" style="font-size: 70.7%; font-family: MathJax_Main;">1</span></span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-231" style="font-family: MathJax_Main; padding-left: 0.238em;">−</span><span class="msub" id="MathJax-Span-232" style="padding-left: 0.238em;"><span style="display: inline-block; position: relative; width: 0.946em; height: 0px;"><span style="position: absolute; clip: rect(2.993em 1000.002em 4.331em -0.313em); top: -4.012em; left: 0.002em;"><span class="mover" id="MathJax-Span-233"><span style="display: inline-block; position: relative; width: 0.632em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.835em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-234" style="font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; clip: rect(1.97em 1000.002em 2.324em -0.274em); top: -2.989em; left: 0.12em;"><span class="mo" id="MathJax-Span-235" style=""><span style="font-family: MathJax_Main;">˜</span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span></span><span style="display: inline-block; width: 0px; height: 4.016em;"></span></span><span style="position: absolute; top: -1.887em; left: 0.632em;"><span class="mi" id="MathJax-Span-236" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-237" style="font-family: MathJax_Main;">)</span><span class="mo" id="MathJax-Span-238" style="font-family: MathJax_Main;">|</span><span class="msub" id="MathJax-Span-239"><span style="display: inline-block; position: relative; width: 1.812em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.835em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-240" style="font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.044em; left: 0.592em;"><span class="mrow" id="MathJax-Span-241"><span class="mi" id="MathJax-Span-242" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span class="mo" id="MathJax-Span-243" style="font-size: 70.7%; font-family: MathJax_Main;">+</span><span class="mn" id="MathJax-Span-244" style="font-size: 70.7%; font-family: MathJax_Main;">1</span></span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-245" style="font-family: MathJax_Main; padding-left: 0.238em;">−</span><span class="msub" id="MathJax-Span-246" style="padding-left: 0.238em;"><span style="display: inline-block; position: relative; width: 0.946em; height: 0px;"><span style="position: absolute; clip: rect(2.993em 1000.002em 4.331em -0.313em); top: -4.012em; left: 0.002em;"><span class="mover" id="MathJax-Span-247"><span style="display: inline-block; position: relative; width: 0.632em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.835em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-248" style="font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; clip: rect(1.97em 1000.002em 2.324em -0.274em); top: -2.989em; left: 0.12em;"><span class="mo" id="MathJax-Span-249" style=""><span style="font-family: MathJax_Main;">˜</span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span></span><span style="display: inline-block; width: 0px; height: 4.016em;"></span></span><span style="position: absolute; top: -1.887em; left: 0.632em;"><span class="mi" id="MathJax-Span-250" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-251" style="font-family: MathJax_Main;">|</span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span><span style="position: absolute; clip: rect(1.891em 1000.002em 3.19em -0.234em); top: -2.084em; left: 50%; margin-left: -1.612em;"><span class="mrow" id="MathJax-Span-252"><span class="mo" id="MathJax-Span-253" style="font-family: MathJax_Main;">|</span><span class="mo" id="MathJax-Span-254" style="font-family: MathJax_Main;">∇</span><span class="msub" id="MathJax-Span-255"><span style="display: inline-block; position: relative; width: 1.812em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.835em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-256" style="font-family: MathJax_Math; font-style: italic;">ϕ</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.044em; left: 0.592em;"><span class="mrow" id="MathJax-Span-257"><span class="mi" id="MathJax-Span-258" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span class="mo" id="MathJax-Span-259" style="font-size: 70.7%; font-family: MathJax_Main;">+</span><span class="mn" id="MathJax-Span-260" style="font-size: 70.7%; font-family: MathJax_Main;">1</span></span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-261" style="font-family: MathJax_Main;">|</span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span><span style="position: absolute; clip: rect(0.868em 1000.002em 1.183em -0.352em); top: -1.297em; left: 0.002em;"><span style="border-left-width: 11.179em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 0.081em; vertical-align: 0.002em;"></span><span style="display: inline-block; width: 0px; height: 1.065em;"></span></span></span></span><span class="mo" id="MathJax-Span-262" style="font-family: MathJax_Main; padding-left: 0.159em;">.</span></span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span><span style="border-left-width: 0.002em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 3.288em; vertical-align: -1.236em;"></span></span></nobr></span></div><script type="math/mml" id="MathJax-Element-7"><math id="M7" display="block" overflow="scroll"><mrow><msub><mi mathvariant="bold">F</mi><mi>C</mi></msub><mo>=</mo><mfrac><mrow><mtext>sign</mtext><mo stretchy="false">(</mo><msub><mi>ϕ</mi><mrow><mi>i</mi><mo>+</mo><mn>1</mn></mrow></msub><mo>−</mo><msub><mover accent="true"><mi>ϕ</mi><mo>~</mo></mover><mi>i</mi></msub><mo stretchy="false">)</mo><mo stretchy="false">|</mo><msub><mi>ϕ</mi><mrow><mi>i</mi><mo>+</mo><mn>1</mn></mrow></msub><mo>−</mo><msub><mover accent="true"><mi>ϕ</mi><mo>~</mo></mover><mi>i</mi></msub><mo stretchy="false">|</mo></mrow><mrow><mo stretchy="false">|</mo><mo>∇</mo><msub><mi>ϕ</mi><mrow><mi>i</mi><mo>+</mo><mn>1</mn></mrow></msub><mo stretchy="false">|</mo></mrow></mfrac><mo>.</mo></mrow></math></script></div><div class="l">(5)</div></div></div><p id="P21">The image correspondence speed <strong>F</strong><sub><em>C</em></sub> can be integrated into the level set <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#FD1" rid="FD1" class=" disp-formula">equation 1</a> like other speed functions. In our implementation, we gradually decrease ξ as the level set iteration proceeds so that the entire active ribbon can move towards the correct location of the target membrane at the beginning, and then becomes more stable at the end such that the ribbon boundaries can close in on the membrane boundaries. <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F4/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F4" rid-ob="ob-F4" co-legend-rid="lgnd_F4"><span>Figure 4</span></a> shows the robust transition of the active ribbon between slices with the image correspondence force.</p><!--fig ft0--><!--fig mode=article f1--><div class="fig iconblock ten_col whole_rhythm clearfix" id="F4" co-legend-rid="lgnd_F4"><a class="icnblk_img figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F4/" target="figure" rid-figpopup="F4" rid-ob="ob-F4"><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/nihms216874f4.gif" class="small-thumb" alt="Fig. 4" title="Fig. 4" src-large="/pmc/articles/PMC3179915/bin/nihms216874f4.jpg"></a><div class="icnblk_cntnt" id="lgnd_F4"><div><a class="figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F4/" target="figure" rid-figpopup="F4" rid-ob="ob-F4">Fig. 4</a></div><!--caption a4--><div><span>Active ribbon with image correspondence force. Left: Input image. Middle left: Segmentation using active ribbon on the current slice. Middle right: Incorrect initial position of active ribbon on the next slice (projection along z-axis). Right: Correct</span> <strong>...</strong></div></div></div></div><div id="S12" class="sec"><h3>4.2 3D Centerline Tracking</h3><p id="P22">To deal with non-axis aligned neural processes, we implemented a tracking algorithm that follows the centerline of the process. Even though tracking a centerline through membrane centers may seem straightforward, it is not simple in our case because we do not know membrane locations in advance. In other words, even though the current slice position and segmentation are given, we do not know the position and segmentation of the next slice.</p><p id="P23">To tackle this problem, we propose a two-step method that consists of estimation and correction steps. In the estimation step, the tangent direction <em>V<sub>t</sub></em> at the last center point is computed using a one-sided finite difference method. We also keep the previous tracking direction <em>V<sub>p</sub></em>. The new tracking direction is then the weighted average between those two vectors: <em>V</em> = ω<em>V<sub>p</sub></em> + (1 − ω)<em>V<sub>t</sub></em> (<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F3/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F3" rid-ob="ob-F3" co-legend-rid="lgnd_F3"><span>Figure 3</span></a> right). The weight ω controls the amount of history used to determine the current tracking direction. We typically use a value of ω = 0.9 for smooth transition between slices.</p><p id="P24">Once we compute a new tracking direction, a temporary new center position <em>C</em><sub><em>i</em>+1</sub> of the next slice can be estimated by simple extrapolation as <em><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/x0043x0303.gif" border="0" alt="C" title=""></em><sub><em>i</em>+1</sub> = <em>C<sub>i</sub></em> + δ<em>V</em>, where δ is the pixel width (i.e., grid spacing) in order to move no more than one pixel distance per estimation step. The local frame of the previous slice is then projected onto the new plane defined by the center <em><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/x0043x0303.gif" border="0" alt="C" title=""></em><sub><em>i</em>+1</sub> and the normal <em>V</em>. A new 2D slice is resampled from the volume data using the new local frame and used for segmentation. Finally, in the correction step, <em><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/x0043x0303.gif" border="0" alt="C" title=""></em><sub><em>i</em>+1</sub> is replaced by the correct center of the segmented neural membrane, <em>C</em><sub><em>i</em>+1</sub>. <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F5/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F5" rid-ob="ob-F5" co-legend-rid="lgnd_F5"><span>Figure 5</span></a> shows an example of 3D centerline tracking and segmentation using NeuroTrace.</p><!--fig ft0--><!--fig mode=article f1--><div class="fig iconblock ten_col whole_rhythm clearfix" id="F5" co-legend-rid="lgnd_F5"><a class="icnblk_img figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F5/" target="figure" rid-figpopup="F5" rid-ob="ob-F5"><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/nihms216874f5.gif" class="small-thumb" alt="Fig. 5" title="Fig. 5" src-large="/pmc/articles/PMC3179915/bin/nihms216874f5.jpg"></a><div class="icnblk_cntnt" id="lgnd_F5"><div><a class="figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F5/" target="figure" rid-figpopup="F5" rid-ob="ob-F5">Fig. 5</a></div><!--caption a4--><div><span>3D segmentation in progress. Green: 2D level set segmentation of neural membranes. Red: 3D centerline tracking.</span></div></div></div></div><div id="S13" class="sec sec-last"><h3>4.3 GPU Implementation</h3><p id="P25">Our GPU level set solver updates the level set only in active regions using a block-based narrow band proposed by Lefohn et al. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R16" rid="R16" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">16</a>]. A slight difference is that we collect all the blocks within a user-defined narrow band size, where the minimum distance to the zero level set of each block is computed in the redistance step without explicitly checking the activation of neighboring blocks. The main level set update process consists of four steps: (1) Form the active list by collecting the active blocks. (2) Iteratively update the level set on each active block in the active list up to the pre-computed number of iterations (based on the narrowband width). (3) Recompute the distance from the zero level set. (4) Stop if the level set converges to a steady state or the maximum number of iterations is reached. Otherwise go to (1).</p><p id="P26">The active list is a one dimensional array of unsigned integers. The first element in this list is the total number of active blocks, and the rest of the array contains the active block indices. To manage the active list efficiently, we store it entirely on the GPU. The only interaction between the CPU and the GPU is copying the first element of the active list from the GPU to the CPU. Then the host code launches a CUDA kernel with the grid size equal to the total number of active blocks. The size of a CUDA block is the same as an active block for our level set. In the CUDA kernel, the global memory address is computed by off-setting from the base address using the active block index. Managing the active list, i.e., adding new active blocks and removing non-active blocks, can be achieved using the atomic hardware operators of recent NVIDIA GPUs without using additional stream compaction processes. We can compute the minimum distance to the zero level set for each block using parallel reduction. If the minimum distance is smaller than the user-defined narrow band width, the total number of active blocks is increased by one using <kbd>AtomicAdd ()</kbd>. Then the current block index is stored at the end of the current list using the index returned by the atomic operator.</p><p id="P27">Once the active list is formed, then each block in the list can be up-dated multiple times depending on the width of the narrow band. For example, if the grid spacing is 1 and the width of the narrow band is 10, then we can safely update the active blocks in the current active list 10 times without refreshing the active list (i.e., explicitly checking the (de-)activation of the blocks). This is because the CourantFriedrichsLewy (CFL) condition [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R26" rid="R26" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">26</a>] guarantees that the maximum deformation incurred by a single update of the level set cannot be greater than the grid spacing. The level set update is done using a Jacobi update method, and communication between block boundaries can be handled implicitly by calling the new CUDA kernel for each level set update because the new solutions are written back to global memory after each update.</p><p id="P28">In extending the single level set method to multiphase level sets we need to evaluate the correct distance between two level sets to keep the topology of the active ribbon consistent. However, the active ribbon does not guarantee the correct distance after deformation due to the combination of various force fields. Therefore, we recompute the distance field for each level set when the list of active blocks is up-dated. Note that we need to redistance not only on the active lists but the complete level sets because the level sets may not share the same active list unless they are very close to each other. To quickly compute the distance fields we employ the GPU-based Eikonal solver by Jeong etal. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R12" rid="R12" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">12</a>].</p><div id="P29"><p>We implemented the nonrigid image registration method using semi-implicit discretization as a two-step iterative process, updating and smoothing the vector field <strong>v</strong> as follows:
245 </p><div class="disp-formula" id="FD8"><div class="f"><div class="MathJax_Display" role="textbox" aria-readonly="true" style="text-align: center;"><span class="MathJax" id="MathJax-Element-8-Frame" style=""><nobr><span class="math" id="M8" style="width: 13.579em; display: inline-block;"><span style="display: inline-block; position: relative; width: 11.218em; height: 0px; font-size: 121%;"><span style="position: absolute; clip: rect(1.773em 1000.002em 3.15em -0.313em); top: -2.792em; left: 0.002em;"><span class="mrow" id="MathJax-Span-264"><span class="mrow" id="MathJax-Span-265"><span class="mi" id="MathJax-Span-266" style="font-family: MathJax_Main; font-weight: bold;">v</span><span class="mo" id="MathJax-Span-267" style="font-family: MathJax_Main; padding-left: 0.277em;">←</span><span class="mi" id="MathJax-Span-268" style="font-family: MathJax_Main; font-weight: bold; padding-left: 0.277em;">v</span><span class="mo" id="MathJax-Span-269" style="font-family: MathJax_Main; padding-left: 0.238em;">+</span><span class="mtext" id="MathJax-Span-270" style="font-family: MathJax_Math; font-style: italic; padding-left: 0.238em;">dt</span><span class="mo" id="MathJax-Span-271" style="font-family: MathJax_Main;">(</span><span class="msub" id="MathJax-Span-272"><span style="display: inline-block; position: relative; width: 1.694em; height: 0px;"><span style="position: absolute; clip: rect(1.734em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-273" style="font-family: MathJax_Math; font-style: italic;">I<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.081em;"></span></span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.123em; left: 0.474em;"><span class="mrow" id="MathJax-Span-274"><span class="mi" id="MathJax-Span-275" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span class="mo" id="MathJax-Span-276" style="font-size: 70.7%; font-family: MathJax_Main;">+</span><span class="mn" id="MathJax-Span-277" style="font-size: 70.7%; font-family: MathJax_Main;">1</span></span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-278" style="font-family: MathJax_Main; padding-left: 0.238em;">−</span><span class="msub" id="MathJax-Span-279" style="padding-left: 0.238em;"><span style="display: inline-block; position: relative; width: 1.025em; height: 0px;"><span style="position: absolute; clip: rect(3.032em 1000.002em 4.134em -0.313em); top: -4.012em; left: 0.002em;"><span class="mover" id="MathJax-Span-280"><span style="display: inline-block; position: relative; width: 0.71em; height: 0px;"><span style="position: absolute; clip: rect(1.734em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-281" style="font-family: MathJax_Math; font-style: italic;">I<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.081em;"></span></span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; clip: rect(1.97em 1000.002em 2.324em -0.274em); top: -2.95em; left: 0.199em;"><span class="mo" id="MathJax-Span-282" style=""><span style="font-family: MathJax_Main;">˜</span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span></span><span style="display: inline-block; width: 0px; height: 4.016em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.71em;"><span class="mi" id="MathJax-Span-283" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-284" style="font-family: MathJax_Main;">)</span><span class="mo" id="MathJax-Span-285" style="font-family: MathJax_Main;">∇</span><span class="msub" id="MathJax-Span-286"><span style="display: inline-block; position: relative; width: 0.828em; height: 0px;"><span style="position: absolute; clip: rect(2.993em 1000.002em 4.134em -0.313em); top: -4.012em; left: 0.002em;"><span class="mover" id="MathJax-Span-287"><span style="display: inline-block; position: relative; width: 0.514em; height: 0px;"><span style="position: absolute; clip: rect(2.009em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.041em;"><span class="mi" id="MathJax-Span-288" style="font-family: MathJax_Main; font-weight: bold;">I</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; clip: rect(1.97em 1000.002em 2.324em -0.274em); top: -2.989em; left: 0.002em;"><span class="mo" id="MathJax-Span-289" style=""><span style="font-family: MathJax_Main;">˜</span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span></span><span style="display: inline-block; width: 0px; height: 4.016em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.514em;"><span class="mi" id="MathJax-Span-290" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span></span><span style="border-left-width: 0.002em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 1.479em; vertical-align: -0.331em;"></span></span></nobr></span></div><script type="math/mml" id="MathJax-Element-8"><math id="M8" display="block" overflow="scroll"><mrow><mi mathvariant="bold">v</mi><mo>←</mo><mi mathvariant="bold">v</mi><mo>+</mo><mtext mathvariant="italic">dt</mtext><mo stretchy="false">(</mo><msub><mi>I</mi><mrow><mi>i</mi><mo>+</mo><mn>1</mn></mrow></msub><mo>−</mo><msub><mover accent="true"><mi>I</mi><mo>~</mo></mover><mi>i</mi></msub><mo stretchy="false">)</mo><mo>∇</mo><msub><mover accent="true"><mi mathvariant="bold">I</mi><mo>~</mo></mover><mi>i</mi></msub></mrow></math></script></div><div class="l">(6)</div></div><p>
247 </p><div class="disp-formula" id="FD9"><div class="f"><div class="MathJax_Display" role="textbox" aria-readonly="true" style="text-align: center;"><span class="MathJax" id="MathJax-Element-9-Frame" style=""><nobr><span class="math" id="M9" style="width: 5.787em; display: inline-block;"><span style="display: inline-block; position: relative; width: 4.764em; height: 0px; font-size: 121%;"><span style="position: absolute; clip: rect(1.93em 1000.002em 3.111em -0.313em); top: -2.792em; left: 0.002em;"><span class="mrow" id="MathJax-Span-292"><span class="mrow" id="MathJax-Span-293"><span class="mi" id="MathJax-Span-294" style="font-family: MathJax_Main; font-weight: bold;">v</span><span class="mo" id="MathJax-Span-295" style="font-family: MathJax_Main; padding-left: 0.277em;">←</span><span class="mi" id="MathJax-Span-296" style="font-family: MathJax_Math; font-style: italic; padding-left: 0.277em;">G</span><span class="mo" id="MathJax-Span-297" style="font-family: MathJax_Main; padding-left: 0.238em;">*</span><span class="mi" id="MathJax-Span-298" style="font-family: MathJax_Main; font-weight: bold; padding-left: 0.238em;">v</span><span class="mo" id="MathJax-Span-299" style="font-family: MathJax_Main;">,</span></span></span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span></span><span style="border-left-width: 0.002em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 1.24em; vertical-align: -0.283em;"></span></span></nobr></span></div><script type="math/mml" id="MathJax-Element-9"><math id="M9" display="block" overflow="scroll"><mrow><mi mathvariant="bold">v</mi><mo>←</mo><mi>G</mi><mo>*</mo><mi mathvariant="bold">v</mi><mo>,</mo></mrow></math></script></div><div class="l">(7)</div></div><p>
248 where <em>G</em> is a Gaussian smoothing kernel. <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#FD8" rid="FD8" class=" disp-formula">Equation 6</a> is a simple Euler integration that can be efficiently mapped to the GPU. To interpolate the pixel values <em>Ĩ<sub>i</sub></em> and <img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/nabla.gif" border="0" alt="[nabla]" title=""><em>Ĩ<sub>i</sub></em> on locations defined by <strong>v</strong> we use texture hardware interpolation on the GPU. Texture memory is cached, so it is efficient for locally coherent random memory accesses. To speed up the 2D Gaussian smoothing in image space, we implemented a sequence of 1D convolutions using shared memory and apply them along x and y, respectively.</p></div></div></div><div id="S14" class="sec"><div class="goto jig-ncbiinpagenav-goto-container"><a class="tgt_dark page-toc-label jig-ncbiinpagenav-goto-heading" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" title="Go to other sections in this page" role="button" aria-expanded="false" aria-haspopup="true">Go to:</a></div><h2 class="head no_bottom_margin ui-helper-clearfix" id="S14title">5 Volume Visualization</h2><p id="P30">Volume rendering of high-resolution EM data poses several challenges. EM data is extremely dense and heavily textured, exhibits a complex structure of interconnected nerve cells, and has a low signal-to-noise ratio. Therefore, standard volume rendering results in cluttered images that make it hard to identify regions of interest (ROIs) or to observe an ongoing segmentation.</p><p id="P31">Our visualization approach supports the inspection of data prior to segmentation, for identifying ROIs, as well as the visualization of the ongoing and final segmentation (see <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F2/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F2" rid-ob="ob-F2" co-legend-rid="lgnd_F2"><span>Figure 2</span></a>). To improve the visualization of the raw data prior to segmentation, we have implemented on-the-fly nonlinear noise removal and edge enhancement to support the user in finding and selecting ROIs. Using a local histogram-based edge metric, which is only calculated on demand for currently visible parts of the volume and cached for later reuse, we can enhance important structures (e.g., myelinated axons) while fading out less important regions. During ray-casting we use the computed edge values to modulate the current sample’s opacity with different user-selectable opacity weighting modes (e.g., min, max, alpha blending).</p><div id="S15" class="sec"><h3>5.1 On-demand Filtering</h3><p id="P32">The main motivations for on-demand filtering (i.e., noise removal and edge detection) are the flexibility offered by being able to change filters and filter parameters on the fly while avoiding additional disk storage and bandwidth bottlenecks for terabyte-sized volume data. We perform filtering only on blocks of the volume that are visible from the current viewpoint, and store the computed data directly on the GPU for later reuse. We have implemented a caching scheme for these pre-computed blocks on the GPU to avoid costly transfers to and from GPU memory while at the same time avoiding repetitive recalculation of filtered blocks. During visualization we display either the original volume, the noise-reduced data, the computed edge values, or a combination of the above.</p><p id="P33">Our on-demand filtering algorithm consists of several steps: (1) Detect for each block in the volume if it is visible from the current viewpoint. (2) Build the list of blocks that need to be computed. (3) Perform noise removal filtering on selected blocks and store them in the cache. (4) Calculate the histogram-based edge metric on selected blocks and store those blocks in the cache. (5) High-resolution ray-casting combining edge values and original data values. The detection of visible blocks (Step 1) is done either in a separate low-resolution ray-casting pass or included in Step 5.</p><div id="S16" class="sec"><p></p><h4 class="inline">5.1.1 Noise Removal </h4><p id="P34">Since EM data generally exhibits a low signal-to-noise ratio we have integrated an on-demand noise removal filter step into our pipeline prior to calculating the local histogram-based edge metric. We perform the filtering only on those blocks that were marked as visible and are not present in the cache yet. We have implemented 2D and 3D Gaussian, mean, non-linear median, bilateral [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R31" rid="R31" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">31</a>], and anisotropic diffusion filters [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R22" rid="R22" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">22</a>] with user adjustable neighborhood sizes. Especially non-linear filters have shown good noise removal properties without degrading edges in the EM data [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R30" rid="R30" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">30</a>]. Our main objective, however, was to develop a general framework for noise removal, where additional filters could be added easily. The results for each processed block is stored in the cache and used as input for the edge detection algorithm.</p></div><div id="S17" class="sec"><p></p><h4 class="inline">5.1.2 Local Histogram-based Edge Detection </h4><p id="P35">We use a local histogram-based edge metric to modulate the opacity of the EM data during raycasting. Boundaries in the volume get enhanced while more homogenous regions are supressed. This helps the user in navigating through the unsegmented dataset and in finding regions where a segmentation should be started. The edge metric is computed only for visible blocks that are not stored in the cache yet.</p><p id="P36">Our edge detection algorithm is based on the work of Martin et al. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R19" rid="R19" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_320459235" role="button" aria-expanded="false" aria-haspopup="true">19</a>] who introduced edge and boundary detection in 2D image based on local histograms. They did a thorough evaluation of different brightness, color, and texture cues for constructing a local boundary model, which was subsequently used to detect contours [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R18" rid="R18" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">18</a>] in natural images.</p><p id="P37">In our local histogram-based edge detection approach we take a block neighborhood around each voxel to calculate the brightness gradient for different directions. We separate the voxel’s neighborhood along the given direction into two halves and calculate the histogram in each half-space. Finally, the histogram difference is calculated using the χ<sup>2</sup> distance metric. A high difference between histograms indicates an abrupt change in brightness in the volume, i.e., an edge. The maximum difference value over all directions is saved as the edge value in the cache block. As the neighborhood size for the histogram calculation can be adjusted to match the resolution level of the current input data, this approach scales to large data and to volume subdivision schemes like octrees. Again, we have kept the implementation of our edge detection framework as modular as possible to support adding different edge detection algorithms in the future. During volume rendering, we fetch at each sample location the corresponding edge value and use it to modulate the sample’s opacity and/or color. Optionally, the user can first use a windowing function on the calculated edge values to further enhance the visualization.</p></div><div id="S18" class="sec"><p></p><h4 class="inline">5.1.3 Dynamic Caching </h4><p id="P38">To improve the performance of our edge-based visualization scheme we have implemented a dynamic caching scheme for storing on-the-fly computed blocks. Two caches are allocated directly on the GPU, one to store de-noised volume blocks and the second to store blocks containing the calculated edge values. First, the visibility of all blocks is updated for the current viewpoint in a first ray-casting pass and saved in a 3D array corresponding to the number of blocks in the volume. Next, all blocks are flagged as either: (1) visible, present in cache; (2) visible, not present in cache; (3) not visible, present in cache; or (4) not visible, not present in cache. Visible blocks that are already in the cache (flagged with (1)) do not need to be recomputed. Only blocks flagged with (2) need to the processed. Therefore, indices of blocks flagged with (2) are stored for later calculation (see Section 5.1.4). During filtering/edge detection the computed blocks are stored in the corresponding cache. A small lookup table is maintained for mapping between block storage space in the cache to actual volume blocks as described in [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R4" rid="R4" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">4</a>]. Unused blocks are kept in the cache for later reuse (flagged with (3)). However, if cache memory gets low, unused blocks are flushed from the cache and replaced by currently visible blocks.</p></div><div id="S19" class="sec sec-last"><p></p><h4 class="inline">5.1.4 GPU Implementation </h4><p id="P39">After detecting which blocks need processing, a CUDA kernel is launched with grid size corresponding to the number of blocks that need to be processed. For simplicity we explain the implementation of our filtering and edge detection algorithm in 2D. The extension to 3D is straightforward.</p><p id="P40">To calculate filter/edge values in each block, we start a CUDA kernel with a CUDA block size that corresponds to the user specified neighborhood size, but with one dimension less than the actual neighborhood (e.g., for a 3D neighborhood a 2D CUDA block is started, for a 2D neighborhood a ID CUDA block is started). <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F6/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F6" rid-ob="ob-F6" co-legend-rid="lgnd_F6"><span>Figure 6</span></a> depicts the case where the edge detection of a block uses a 5×5 neighborhood. In this case the kernel is started with 5 concurrent threads. Next, the threads iterate over the entire block that needs to be filtered and calculate the filter/edge values for each voxel. Each thread is responsible for only one part of the filter’s neighborhood, as depicted by the colored areas in <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F6/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F6" rid-ob="ob-F6" co-legend-rid="lgnd_F6"><span>Figure 6</span></a>. To reduce redundant texture fetches each thread locally caches its last computed values. The size of this thread-local array corresponds to the neighborhood size of the filter. Therefore, at each step a thread only needs to perform one texture fetch, and store the value in its local cache (<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F6/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F6" rid-ob="ob-F6" co-legend-rid="lgnd_F6"><span>Figure 6</span></a>, middle).</p><!--fig ft0--><!--fig mode=article f1--><div class="fig iconblock ten_col whole_rhythm clearfix" id="F6" co-legend-rid="lgnd_F6"><a class="icnblk_img figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F6/" target="figure" rid-figpopup="F6" rid-ob="ob-F6"><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/nihms216874f6.gif" class="small-thumb" alt="Fig. 6" title="Fig. 6" src-large="/pmc/articles/PMC3179915/bin/nihms216874f6.jpg"></a><div class="icnblk_cntnt" id="lgnd_F6"><div><a class="figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F6/" target="figure" rid-figpopup="F6" rid-ob="ob-F6">Fig. 6</a></div><!--caption a4--><div><span>Local histogram-based edge detection in volume blocks using CUDA. Left: Neigbhorhood required for local histograms. Center: Fetching only one new sample per thread at each step to update the neighborhood. Right: Shared histograms for calculation of the</span> <strong>...</strong></div></div></div><p id="P41">To calculate the local-histogram based edge metric, all samples in a voxel’s neighborhood need to be assigned to one of the two local histograms (for both half-spaces), as depicted in <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F6/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F6" rid-ob="ob-F6" co-legend-rid="lgnd_F6"><span>Figure 6</span></a>, right. The histograms are stored in shared CUDA memory and used for the final calculation of the χ<sup>2</sup> histogram difference. The main steps for each thread are: (1) Update the histogram of the first half-space (<em>histogram<sub>left</sub></em>) by removing the sample that has left the filter neighborhood and adding the last sample from <em>histogram<sub>right</sub></em>. (2) Remove the sample that has left the filter neighborhood from the thread-local cache. (3) Fetch the sample that has entered the filter neighborhood from the volume texture and store it in the thread-local cache. (4) Update the histogram of the second half-space (<em>histogram<sub>right</sub></em>) by removing the sample that is now in <em>histogram<sub>left</sub></em> and adding the sample that has just been fetched from the volume texture. All threads are synchronized after they have performed the above steps using atomic CUDA operations for updating the shared histograms. Now the χ<sup>2</sup> histogram difference for the current neighborhood can be computed and stored in the cache.</p><p id="P42">To implement the de-noising filters we use the same basic strategy. For Gaussian filters we transfer a 1D look-up table of the weights to the GPU to speed up the calculation. For bilateral filtering we use the same look-up table to calculate the geometric closeness function, whereas the photometric similarity function is calculated on-the-fly in the CUDA kernel. For median filtering we implemented bitonic sort on the GPU to find the median value of the filter neighborhood. Anisotropic diffusion filtering is the most complex filter in our framework. It requires a second filter cache to allow ping-pong swaps between source and destination. Also, costly neighborhood lookups in the source cache are needed to compute the boundary values of the destination blocks.</p><p id="P43">If the noise removal step is performed prior to the edge detection, the local histogram calculation uses the values from the filtered block cache as input values instead of the original volume texture. Therefore, special care has to be taken when fetching de-noised values for a neighborhood at an edge-block’s boundary. This case can be handled by either extending the dimensions of the de-noised blocks compared to the edge-detected blocks, or by detecting which additional blocks would have to be de-noised and performing a neighborhood lookup for areas outside an edge block’ boundary.</p></div></div><div id="S20" class="sec sec-last"><h3>5.2 Visualization of Segmented Neural Structures</h3><p id="P44">In order to visualize and inspect the segmented neural processes in 3D, we depict the original volume data together with semi-transparent iso-surfaces that delineate structures such as axons or dendrites. The output image is generated in a single ray-casting pass for both the iso-surfaces and the part of the volume that is shown using direct volume rendering. While stepping from sample to sample along a given viewing ray, the direct volume rendering integral is solved via front-to-back compositing. At the same time, each sample is tested for potential intersections with iso-surfaces. If a surface is intersected, its color and transparency are composited with the accumulated volume-rendered part, and direct volume rendering is continued behind the surface intersection.</p><p id="P45">Our active ribbon segmentation described in Section 4 outputs a set of implicit surfaces for each 2D slice. However, in order to make the system scalable for large EM data, we do not store these 2D distance fields. Instead, we convert the segmentation to a very compact format by fitting an ellipse to each active ribbon. An entire structure such as an axon is then represented as a simple list of elliptical cross-sections, which reduces the memory footprint significantly.</p><div id="S21" class="sec"><p></p><h4 class="inline">5.2.1 Implicit Surfaces from Elliptical Cross-Sections </h4><p id="P46">To render smooth, connected surfaces from elliptical cross-sections we compute implicit surfaces from the set of ellipses on-the-fly. Although everything is computed in a single ray-casting pass, in order to simplify the problem we treat it as two conceptually separate parts. The first part is ray-casting of an implicit surface in a distance field ϕ(<strong>x</strong>), where the surface is defined by the points where ϕ(<strong>x</strong>) = 0. The second part is the computation of ϕ(<strong>x</strong>) for any point <strong>x</strong> in volume space. Here, a point <strong>x</strong> is either a sample <strong>p</strong> on a viewing ray, or the location of a central differences computation during shading.</p><p id="P47">The ray-caster renders and shades implicit surfaces by evaluating ϕ(<strong>p</strong>) along viewing rays. Intersection with an implicit surface is detected between two successive points on a ray when ϕ(<strong>p</strong><sub><em>i</em></sub>) < 0 and ϕ(<strong>p</strong><sub><em>i</em>+1</sub>) > 0, i.e., the first sample is in front and the second one behind the surface. We carry out a predetermined number of bisection steps in order to find a sufficiently accurate intersection position [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R10" rid="R10" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">10</a>]. At that location, we compute the shading using the normalized gradient obtained from central differences in the distance field, evaluating ϕ(<strong>x</strong>) at six additional locations. In order to simplify accommodating multiple axons that are represented as implicit surfaces each, we map every point <strong>x</strong> in volume space for surface intersection and shading purposes to only one distance ϕ(<strong>x</strong>), which is the distance to the closest axon. Conceptually, each axon is represented by a 3D distance field ϕ<sub><em>i</em></sub>, and the ϕ(<strong>x</strong>) used in the ray-casting loop is ϕ(<strong>x</strong>) = min<sub><em>i</em></sub> (ϕ<sub><em>i</em></sub>(<strong>x</strong>)).</p><p id="P48">In order to obtain implicit surfaces in 3D from a collection of 2D elliptical cross-sections, we have to be able to evaluate ϕ(<strong>x</strong>) throughout the volume. We want this interpolation to be fast and easy to implement, and still result in smooth shaded surfaces. Each cross-section is represented by a single ellipse <strong>ell</strong><sub><em>i</em></sub> = (<strong>c</strong><sub><em>i</em></sub>,<strong>q</strong><sub><em>i</em></sub>, <span class="MathJax" id="MathJax-Element-10-Frame" role="textbox" aria-readonly="true" style=""><nobr><span class="math" id="M10" style="width: 2.333em; display: inline-block;"><span style="display: inline-block; position: relative; width: 1.9em; height: 0px; font-size: 123%;"><span style="position: absolute; clip: rect(1.791em 1000.003em 3.255em -0.431em); top: -2.762em; left: 0.003em;"><span class="mrow" id="MathJax-Span-301"><span class="mrow" id="MathJax-Span-302"><span class="msubsup" id="MathJax-Span-303"><span style="display: inline-block; position: relative; width: 0.707em; height: 0px;"><span style="position: absolute; clip: rect(1.683em 1000.003em 2.713em -0.431em); top: -2.545em; left: 0.003em;"><span class="mi" id="MathJax-Span-304" style="font-family: MathJax_Math; font-style: italic;">l</span><span style="display: inline-block; width: 0px; height: 2.55em;"></span></span><span style="position: absolute; clip: rect(1.575em 1000.003em 2.225em -0.485em); top: -2.491em; left: 0.274em;"><span class="mi" id="MathJax-Span-305" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">x</span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span><span style="position: absolute; clip: rect(1.412em 1000.003em 2.225em -0.485em); top: -1.786em; left: 0.274em;"><span class="mi" id="MathJax-Span-306" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span></span></span><span class="mo" id="MathJax-Span-307" style="font-family: MathJax_Main;">,</span><span class="msubsup" id="MathJax-Span-308" style="padding-left: 0.165em;"><span style="display: inline-block; position: relative; width: 0.707em; height: 0px;"><span style="position: absolute; clip: rect(1.683em 1000.003em 2.713em -0.431em); top: -2.545em; left: 0.003em;"><span class="mi" id="MathJax-Span-309" style="font-family: MathJax_Math; font-style: italic;">l</span><span style="display: inline-block; width: 0px; height: 2.55em;"></span></span><span style="position: absolute; clip: rect(1.575em 1000.003em 2.388em -0.485em); top: -2.545em; left: 0.274em;"><span class="mi" id="MathJax-Span-310" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">y<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.003em;"></span></span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span><span style="position: absolute; clip: rect(1.412em 1000.003em 2.225em -0.485em); top: -1.732em; left: 0.274em;"><span class="mi" id="MathJax-Span-311" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.767em;"></span></span></span><span style="border-left-width: 0.003em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 1.537em; vertical-align: -0.463em;"></span></span></nobr></span><script type="math/mml" id="MathJax-Element-10"><math id="M10" overflow="scroll"><mrow><msubsup><mi>l</mi><mi>i</mi><mi>x</mi></msubsup><mo>,</mo><msubsup><mi>l</mi><mi>i</mi><mi>y</mi></msubsup></mrow></math></script>), where <strong>c</strong><sub><em>i</em></sub> is the center in 3D, and the 3D coordinate frame is represented by the unit quaternion <strong>q</strong><sub><em>i</em></sub> and the lengths of the ellipse’s semi-axes, <span class="MathJax" id="MathJax-Element-11-Frame" role="textbox" aria-readonly="true" style=""><nobr><span class="math" id="M11" style="width: 4.393em; display: inline-block;"><span style="display: inline-block; position: relative; width: 3.58em; height: 0px; font-size: 123%;"><span style="position: absolute; clip: rect(1.791em 1000.003em 3.255em -0.431em); top: -2.762em; left: 0.003em;"><span class="mrow" id="MathJax-Span-313"><span class="mrow" id="MathJax-Span-314"><span class="msubsup" id="MathJax-Span-315"><span style="display: inline-block; position: relative; width: 0.707em; height: 0px;"><span style="position: absolute; clip: rect(1.683em 1000.003em 2.713em -0.431em); top: -2.545em; left: 0.003em;"><span class="mi" id="MathJax-Span-316" style="font-family: MathJax_Math; font-style: italic;">l</span><span style="display: inline-block; width: 0px; height: 2.55em;"></span></span><span style="position: absolute; clip: rect(1.575em 1000.003em 2.225em -0.485em); top: -2.491em; left: 0.274em;"><span class="mi" id="MathJax-Span-317" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">x</span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span><span style="position: absolute; clip: rect(1.412em 1000.003em 2.225em -0.485em); top: -1.786em; left: 0.274em;"><span class="mi" id="MathJax-Span-318" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span></span></span><span class="mtext" id="MathJax-Span-319" style="font-family: MathJax_Main;"> and </span><span class="msubsup" id="MathJax-Span-320"><span style="display: inline-block; position: relative; width: 0.707em; height: 0px;"><span style="position: absolute; clip: rect(1.683em 1000.003em 2.713em -0.431em); top: -2.545em; left: 0.003em;"><span class="mi" id="MathJax-Span-321" style="font-family: MathJax_Math; font-style: italic;">l</span><span style="display: inline-block; width: 0px; height: 2.55em;"></span></span><span style="position: absolute; clip: rect(1.575em 1000.003em 2.388em -0.485em); top: -2.545em; left: 0.274em;"><span class="mi" id="MathJax-Span-322" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">y<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.003em;"></span></span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span><span style="position: absolute; clip: rect(1.412em 1000.003em 2.225em -0.485em); top: -1.732em; left: 0.274em;"><span class="mi" id="MathJax-Span-323" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.767em;"></span></span></span><span style="border-left-width: 0.003em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 1.537em; vertical-align: -0.463em;"></span></span></nobr></span><script type="math/mml" id="MathJax-Element-11"><math id="M11" overflow="scroll"><mrow><msubsup><mi>l</mi><mi>i</mi><mi>x</mi></msubsup><mtext> and </mtext><msubsup><mi>l</mi><mi>i</mi><mi>y</mi></msubsup></mrow></math></script>, respectively. This results in a very compact representation with just nine floating point values per ellipse. During rendering, we convert between quaternions and explicit coordinate frames when needed, denoting the normal vector of the ellipse’s plane as <strong>n</strong><sub><em>i</em></sub>. Additionally, an integer axon ID is stored with each ellipse, which allows rendering axons with individual color and transparency.</p><div id="P49"><p>The main idea for evaluating ϕ(<strong>p</strong>) at a given point <strong>p</strong> is to directly compute a single interpolated ellipse whose plane exactly or almost contains <strong>p</strong>, and then to perform a straight-forward 2D point-to-ellipse distance computation in this plane. That is, we first compute an interpolated ellipse <strong>ell</strong>(<strong>p</strong>) = (<strong>c</strong><sub><em>p</em></sub>,<strong>q</strong><sub><em>p</em></sub>, <span class="MathJax" id="MathJax-Element-12-Frame" role="textbox" aria-readonly="true" style=""><nobr><span class="math" id="M12" style="width: 2.333em; display: inline-block;"><span style="display: inline-block; position: relative; width: 1.9em; height: 0px; font-size: 123%;"><span style="position: absolute; clip: rect(1.791em 1000.003em 3.255em -0.431em); top: -2.762em; left: 0.003em;"><span class="mrow" id="MathJax-Span-325"><span class="mrow" id="MathJax-Span-326"><span class="msubsup" id="MathJax-Span-327"><span style="display: inline-block; position: relative; width: 0.707em; height: 0px;"><span style="position: absolute; clip: rect(1.683em 1000.003em 2.713em -0.431em); top: -2.545em; left: 0.003em;"><span class="mi" id="MathJax-Span-328" style="font-family: MathJax_Math; font-style: italic;">l</span><span style="display: inline-block; width: 0px; height: 2.55em;"></span></span><span style="position: absolute; clip: rect(1.575em 1000.003em 2.225em -0.485em); top: -2.491em; left: 0.274em;"><span class="mi" id="MathJax-Span-329" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">x</span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span><span style="position: absolute; clip: rect(1.412em 1000.003em 2.225em -0.485em); top: -1.786em; left: 0.274em;"><span class="mi" id="MathJax-Span-330" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span></span></span><span class="mo" id="MathJax-Span-331" style="font-family: MathJax_Main;">,</span><span class="msubsup" id="MathJax-Span-332" style="padding-left: 0.165em;"><span style="display: inline-block; position: relative; width: 0.707em; height: 0px;"><span style="position: absolute; clip: rect(1.683em 1000.003em 2.713em -0.431em); top: -2.545em; left: 0.003em;"><span class="mi" id="MathJax-Span-333" style="font-family: MathJax_Math; font-style: italic;">l</span><span style="display: inline-block; width: 0px; height: 2.55em;"></span></span><span style="position: absolute; clip: rect(1.575em 1000.003em 2.388em -0.485em); top: -2.545em; left: 0.274em;"><span class="mi" id="MathJax-Span-334" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">y<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.003em;"></span></span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span><span style="position: absolute; clip: rect(1.412em 1000.003em 2.225em -0.485em); top: -1.732em; left: 0.274em;"><span class="mi" id="MathJax-Span-335" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.767em;"></span></span></span><span style="border-left-width: 0.003em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 1.537em; vertical-align: -0.463em;"></span></span></nobr></span><script type="math/mml" id="MathJax-Element-12"><math id="M12" overflow="scroll"><mrow><msubsup><mi>l</mi><mi>i</mi><mi>x</mi></msubsup><mo>,</mo><msubsup><mi>l</mi><mi>i</mi><mi>y</mi></msubsup></mrow></math></script>), whose plane (<strong>c</strong><sub><em>p</em></sub>,<strong>n</strong><sub><em>p</em></sub>) approximately contains <strong>p</strong>, then obtain a point <strong>p</strong>′ by projecting <strong>p</strong> into this plane, and finally compute the distance of <strong>p</strong>′ to <strong>ell</strong>(<strong>p</strong>) in 2D. The distance for <strong>p</strong> is thus approximated as ϕ(<strong>p</strong>) ≈ ϕ(<strong>ell</strong>(<strong>p</strong>),<strong>p</strong>′). The ellipse <strong>ell</strong>(<strong>p</strong>) is interpolated between the nearest pair of ellipses (<strong>ell</strong><sub><em>i</em></sub>,<strong>ell</strong><sub><em>i</em>+1</sub>) that encloses <strong>p</strong>. This pair is the one where <strong>p</strong> is in the front halfspace of <strong>ell</strong><sub><em>i</em></sub>, i.e., <strong>p</strong> · <strong>n</strong><sub><em>i</em></sub> > <strong>c</strong><sub><em>i</em></sub> · <strong>n</strong><sub><em>i</em></sub>, and the back halfspace of <strong>ell</strong><sub><em>i</em>+1</sub>, i.e., <strong>p</strong> · <strong>n</strong><sub><em>i</em>+1</sub> < <strong>c</strong><sub><em>i</em>+1</sub> · <strong>n</strong><sub><em>i</em>+1</sub>. For interpolation, a parameter α <img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/x2208.gif" border="0" alt="[set membership]" title=""> [0, 1] is required for a given <strong>p</strong>, which we compute as follows:
249 </p><div class="disp-formula" id="FD10"><div class="f"><div class="MathJax_Display" role="textbox" aria-readonly="true" style="text-align: center;"><span class="MathJax" id="MathJax-Element-13-Frame" style=""><nobr><span class="math" id="M13" style="width: 35.618em; display: inline-block;"><span style="display: inline-block; position: relative; width: 29.439em; height: 0px; font-size: 121%;"><span style="position: absolute; clip: rect(1.261em 1000.002em 3.78em -0.313em); top: -2.753em; left: 0.002em;"><span class="mrow" id="MathJax-Span-337"><span class="mrow" id="MathJax-Span-338"><span class="mi" id="MathJax-Span-339" style="font-family: MathJax_Math; font-style: italic;">α</span><span class="mo" id="MathJax-Span-340" style="font-family: MathJax_Main; padding-left: 0.277em;">=</span><span class="mfrac" id="MathJax-Span-341" style="padding-left: 0.396em; padding-right: 0.12em;"><span style="display: inline-block; position: relative; width: 3.268em; height: 0px;"><span style="position: absolute; clip: rect(3.19em 1000.002em 4.292em -0.313em); top: -4.681em; left: 50%; margin-left: -0.47em;"><span class="mrow" id="MathJax-Span-342"><span class="msub" id="MathJax-Span-343"><span style="display: inline-block; position: relative; width: 0.946em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-344" style="font-family: MathJax_Math; font-style: italic;">k</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.123em; left: 0.514em;"><span class="mn" id="MathJax-Span-345" style="font-size: 70.7%; font-family: MathJax_Main;">0</span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 4.016em;"></span></span><span style="position: absolute; clip: rect(1.93em 1000.002em 3.032em -0.313em); top: -2.084em; left: 50%; margin-left: -1.572em;"><span class="mrow" id="MathJax-Span-346"><span class="msub" id="MathJax-Span-347"><span style="display: inline-block; position: relative; width: 0.946em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-348" style="font-family: MathJax_Math; font-style: italic;">k</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.123em; left: 0.514em;"><span class="mn" id="MathJax-Span-349" style="font-size: 70.7%; font-family: MathJax_Main;">0</span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-350" style="font-family: MathJax_Main; padding-left: 0.238em;">+</span><span class="msub" id="MathJax-Span-351" style="padding-left: 0.238em;"><span style="display: inline-block; position: relative; width: 0.946em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-352" style="font-family: MathJax_Math; font-style: italic;">k</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.123em; left: 0.514em;"><span class="mn" id="MathJax-Span-353" style="font-size: 70.7%; font-family: MathJax_Main;">1</span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span><span style="position: absolute; clip: rect(0.868em 1000.002em 1.183em -0.352em); top: -1.297em; left: 0.002em;"><span style="border-left-width: 3.268em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 0.081em; vertical-align: 0.002em;"></span><span style="display: inline-block; width: 0px; height: 1.065em;"></span></span></span></span><span class="mtext" id="MathJax-Span-354" style="font-family: MathJax_Main; padding-left: 0.159em;"> with </span><span class="msub" id="MathJax-Span-355"><span style="display: inline-block; position: relative; width: 0.946em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-356" style="font-family: MathJax_Math; font-style: italic;">k</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.123em; left: 0.514em;"><span class="mn" id="MathJax-Span-357" style="font-size: 70.7%; font-family: MathJax_Main;">0</span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-358" style="font-family: MathJax_Main; padding-left: 0.277em;">=</span><span class="mfrac" id="MathJax-Span-359" style="padding-left: 0.396em; padding-right: 0.12em;"><span style="display: inline-block; position: relative; width: 2.56em; height: 0px;"><span style="position: absolute; clip: rect(2.245em 1000.002em 3.111em -0.313em); top: -3.461em; left: 50%; margin-left: -1.179em;"><span class="mrow" id="MathJax-Span-360"><span class="mi" id="MathJax-Span-361" style="font-family: MathJax_Main; font-weight: bold;">p</span><span class="mo" id="MathJax-Span-362" style="font-family: MathJax_Main; padding-left: 0.238em;">⋅</span><span class="msub" id="MathJax-Span-363" style="padding-left: 0.238em;"><span style="display: inline-block; position: relative; width: 0.946em; height: 0px;"><span style="position: absolute; clip: rect(2.245em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.002em;"><span class="mi" id="MathJax-Span-364" style="font-family: MathJax_Main; font-weight: bold;">n</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.632em;"><span class="mi" id="MathJax-Span-365" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; clip: rect(2.048em 1000.002em 3.032em -0.352em); top: -2.084em; left: 50%; margin-left: -1.218em;"><span class="mrow" id="MathJax-Span-366"><span class="mover" id="MathJax-Span-367"><span style="display: inline-block; position: relative; width: 0.71em; height: 0px;"><span style="position: absolute; clip: rect(2.245em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.041em;"><span class="mi" id="MathJax-Span-368" style="font-family: MathJax_Main; font-weight: bold;">n</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; clip: rect(2.048em 1000.002em 2.324em -0.352em); top: -2.753em; left: 0.002em;"><span class="mo" id="MathJax-Span-369" style=""><span style="display: inline-block; position: relative; width: 0.71em; height: 0px;"><span style="position: absolute; font-family: MathJax_Main; top: -4.012em; left: -0.077em;">¯<span style="display: inline-block; width: 0px; height: 4.016em;"></span></span><span style="position: absolute; font-family: MathJax_Main; top: -4.012em; left: 0.277em;">¯<span style="display: inline-block; width: 0px; height: 4.016em;"></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span></span><span class="mo" id="MathJax-Span-370" style="font-family: MathJax_Main; padding-left: 0.238em;">⋅</span><span class="msub" id="MathJax-Span-371" style="padding-left: 0.238em;"><span style="display: inline-block; position: relative; width: 0.946em; height: 0px;"><span style="position: absolute; clip: rect(2.245em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.002em;"><span class="mi" id="MathJax-Span-372" style="font-family: MathJax_Main; font-weight: bold;">n</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.632em;"><span class="mi" id="MathJax-Span-373" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span><span style="position: absolute; clip: rect(0.868em 1000.002em 1.183em -0.352em); top: -1.297em; left: 0.002em;"><span style="border-left-width: 2.56em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 0.081em; vertical-align: 0.002em;"></span><span style="display: inline-block; width: 0px; height: 1.065em;"></span></span></span></span><span class="mo" id="MathJax-Span-374" style="font-family: MathJax_Main; padding-left: 0.159em;">,</span><span class="msub" id="MathJax-Span-375" style="padding-left: 0.159em;"><span style="display: inline-block; position: relative; width: 0.946em; height: 0px;"><span style="position: absolute; clip: rect(1.694em 1000.002em 2.639em -0.313em); top: -2.517em; left: 0.002em;"><span class="mi" id="MathJax-Span-376" style="font-family: MathJax_Math; font-style: italic;">k</span><span style="display: inline-block; width: 0px; height: 2.521em;"></span></span><span style="position: absolute; top: -2.123em; left: 0.514em;"><span class="mn" id="MathJax-Span-377" style="font-size: 70.7%; font-family: MathJax_Main;">1</span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span><span class="mo" id="MathJax-Span-378" style="font-family: MathJax_Main; padding-left: 0.277em;">=</span><span class="mo" id="MathJax-Span-379" style="font-family: MathJax_Main; padding-left: 0.277em;">−</span><span class="mfrac" id="MathJax-Span-380" style="padding-left: 0.277em; padding-right: 0.12em;"><span style="display: inline-block; position: relative; width: 3.465em; height: 0px;"><span style="position: absolute; clip: rect(2.245em 1000.002em 3.111em -0.313em); top: -3.461em; left: 50%; margin-left: -1.612em;"><span class="mrow" id="MathJax-Span-381"><span class="mi" id="MathJax-Span-382" style="font-family: MathJax_Main; font-weight: bold;">p</span><span class="mo" id="MathJax-Span-383" style="font-family: MathJax_Main; padding-left: 0.238em;">⋅</span><span class="msub" id="MathJax-Span-384" style="padding-left: 0.238em;"><span style="display: inline-block; position: relative; width: 1.852em; height: 0px;"><span style="position: absolute; clip: rect(2.245em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.002em;"><span class="mi" id="MathJax-Span-385" style="font-family: MathJax_Main; font-weight: bold;">n</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; top: -2.123em; left: 0.632em;"><span class="mrow" id="MathJax-Span-386"><span class="mi" id="MathJax-Span-387" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span class="mo" id="MathJax-Span-388" style="font-size: 70.7%; font-family: MathJax_Main;">+</span><span class="mn" id="MathJax-Span-389" style="font-size: 70.7%; font-family: MathJax_Main;">1</span></span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; clip: rect(2.048em 1000.002em 3.072em -0.352em); top: -2.084em; left: 50%; margin-left: -1.651em;"><span class="mrow" id="MathJax-Span-390"><span class="mover" id="MathJax-Span-391"><span style="display: inline-block; position: relative; width: 0.71em; height: 0px;"><span style="position: absolute; clip: rect(2.245em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.041em;"><span class="mi" id="MathJax-Span-392" style="font-family: MathJax_Main; font-weight: bold;">n</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; clip: rect(2.048em 1000.002em 2.324em -0.352em); top: -2.753em; left: 0.002em;"><span class="mo" id="MathJax-Span-393" style=""><span style="display: inline-block; position: relative; width: 0.71em; height: 0px;"><span style="position: absolute; font-family: MathJax_Main; top: -4.012em; left: -0.077em;">¯<span style="display: inline-block; width: 0px; height: 4.016em;"></span></span><span style="position: absolute; font-family: MathJax_Main; top: -4.012em; left: 0.277em;">¯<span style="display: inline-block; width: 0px; height: 4.016em;"></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span></span><span class="mo" id="MathJax-Span-394" style="font-family: MathJax_Main; padding-left: 0.238em;">⋅</span><span class="msub" id="MathJax-Span-395" style="padding-left: 0.238em;"><span style="display: inline-block; position: relative; width: 1.852em; height: 0px;"><span style="position: absolute; clip: rect(2.245em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.002em;"><span class="mi" id="MathJax-Span-396" style="font-family: MathJax_Main; font-weight: bold;">n</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; top: -2.123em; left: 0.632em;"><span class="mrow" id="MathJax-Span-397"><span class="mi" id="MathJax-Span-398" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span class="mo" id="MathJax-Span-399" style="font-size: 70.7%; font-family: MathJax_Main;">+</span><span class="mn" id="MathJax-Span-400" style="font-size: 70.7%; font-family: MathJax_Main;">1</span></span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span><span style="position: absolute; clip: rect(0.868em 1000.002em 1.183em -0.352em); top: -1.297em; left: 0.002em;"><span style="border-left-width: 3.465em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 0.081em; vertical-align: 0.002em;"></span><span style="display: inline-block; width: 0px; height: 1.065em;"></span></span></span></span><span class="mo" id="MathJax-Span-401" style="font-family: MathJax_Main; padding-left: 0.159em;">,</span><span class="mover" id="MathJax-Span-402" style="padding-left: 0.159em;"><span style="display: inline-block; position: relative; width: 0.71em; height: 0px;"><span style="position: absolute; clip: rect(2.245em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.041em;"><span class="mi" id="MathJax-Span-403" style="font-family: MathJax_Main; font-weight: bold;">n</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; clip: rect(2.048em 1000.002em 2.324em -0.352em); top: -2.753em; left: 0.002em;"><span class="mo" id="MathJax-Span-404" style=""><span style="display: inline-block; position: relative; width: 0.71em; height: 0px;"><span style="position: absolute; font-family: MathJax_Main; top: -4.012em; left: -0.077em;">¯<span style="display: inline-block; width: 0px; height: 4.016em;"></span></span><span style="position: absolute; font-family: MathJax_Main; top: -4.012em; left: 0.277em;">¯<span style="display: inline-block; width: 0px; height: 4.016em;"></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span></span><span class="mo" id="MathJax-Span-405" style="font-family: MathJax_Main; padding-left: 0.277em;">=</span><span class="mfrac" id="MathJax-Span-406" style="padding-left: 0.396em; padding-right: 0.12em;"><span style="display: inline-block; position: relative; width: 5.669em; height: 0px;"><span style="position: absolute; clip: rect(2.048em 1000.002em 3.072em -0.313em); top: -3.422em; left: 50%; margin-left: -2.044em;"><span class="mrow" id="MathJax-Span-407"><span class="msub" id="MathJax-Span-408"><span style="display: inline-block; position: relative; width: 0.946em; height: 0px;"><span style="position: absolute; clip: rect(2.245em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.002em;"><span class="mi" id="MathJax-Span-409" style="font-family: MathJax_Main; font-weight: bold;">n</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.632em;"><span class="mi" id="MathJax-Span-410" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-411" style="font-family: MathJax_Main; padding-left: 0.238em;">+</span><span class="msub" id="MathJax-Span-412" style="padding-left: 0.238em;"><span style="display: inline-block; position: relative; width: 1.852em; height: 0px;"><span style="position: absolute; clip: rect(2.245em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.002em;"><span class="mi" id="MathJax-Span-413" style="font-family: MathJax_Main; font-weight: bold;">n</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; top: -2.123em; left: 0.632em;"><span class="mrow" id="MathJax-Span-414"><span class="mi" id="MathJax-Span-415" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span class="mo" id="MathJax-Span-416" style="font-size: 70.7%; font-family: MathJax_Main;">+</span><span class="mn" id="MathJax-Span-417" style="font-size: 70.7%; font-family: MathJax_Main;">1</span></span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span><span style="position: absolute; clip: rect(2.127em 1000.002em 3.15em -0.077em); top: -2.163em; left: 50%; margin-left: -2.753em;"><span class="mrow" id="MathJax-Span-418"><span class="mo" id="MathJax-Span-419" style="font-family: MathJax_Size1;">‖</span><span class="mrow" id="MathJax-Span-420"><span class="msub" id="MathJax-Span-421"><span style="display: inline-block; position: relative; width: 0.946em; height: 0px;"><span style="position: absolute; clip: rect(2.245em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.002em;"><span class="mi" id="MathJax-Span-422" style="font-family: MathJax_Main; font-weight: bold;">n</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; top: -1.966em; left: 0.632em;"><span class="mi" id="MathJax-Span-423" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.127em;"></span></span></span></span><span class="mo" id="MathJax-Span-424" style="font-family: MathJax_Main; padding-left: 0.238em;">+</span><span class="msub" id="MathJax-Span-425" style="padding-left: 0.238em;"><span style="display: inline-block; position: relative; width: 1.852em; height: 0px;"><span style="position: absolute; clip: rect(2.245em 1000.002em 2.914em -0.313em); top: -2.792em; left: 0.002em;"><span class="mi" id="MathJax-Span-426" style="font-family: MathJax_Main; font-weight: bold;">n</span><span style="display: inline-block; width: 0px; height: 2.796em;"></span></span><span style="position: absolute; top: -2.123em; left: 0.632em;"><span class="mrow" id="MathJax-Span-427"><span class="mi" id="MathJax-Span-428" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span class="mo" id="MathJax-Span-429" style="font-size: 70.7%; font-family: MathJax_Main;">+</span><span class="mn" id="MathJax-Span-430" style="font-size: 70.7%; font-family: MathJax_Main;">1</span></span><span style="display: inline-block; width: 0px; height: 2.285em;"></span></span></span></span></span><span class="mo" id="MathJax-Span-431" style="font-family: MathJax_Size1;">‖</span></span><span style="display: inline-block; width: 0px; height: 2.835em;"></span></span><span style="position: absolute; clip: rect(0.868em 1000.002em 1.183em -0.352em); top: -1.297em; left: 0.002em;"><span style="border-left-width: 5.669em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 0.081em; vertical-align: 0.002em;"></span><span style="display: inline-block; width: 0px; height: 1.065em;"></span></span></span></span><span class="mo" id="MathJax-Span-432" style="font-family: MathJax_Main; padding-left: 0.159em;">.</span></span></span><span style="display: inline-block; width: 0px; height: 2.757em;"></span></span></span><span style="border-left-width: 0.002em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 2.86em; vertical-align: -1.14em;"></span></span></nobr></span></div><script type="math/mml" id="MathJax-Element-13"><math id="M13" display="block" overflow="scroll"><mrow><mi>α</mi><mo>=</mo><mfrac><mrow><msub><mi>k</mi><mn>0</mn></msub></mrow><mrow><msub><mi>k</mi><mn>0</mn></msub><mo>+</mo><msub><mi>k</mi><mn>1</mn></msub></mrow></mfrac><mtext> with </mtext><msub><mi>k</mi><mn>0</mn></msub><mo>=</mo><mfrac><mrow><mi mathvariant="bold">p</mi><mo>·</mo><msub><mi mathvariant="bold">n</mi><mi>i</mi></msub></mrow><mrow><mover accent="true"><mi mathvariant="bold">n</mi><mo>¯</mo></mover><mo>·</mo><msub><mi mathvariant="bold">n</mi><mi>i</mi></msub></mrow></mfrac><mo>,</mo><msub><mi>k</mi><mn>1</mn></msub><mo>=</mo><mo>−</mo><mfrac><mrow><mi mathvariant="bold">p</mi><mo>·</mo><msub><mi mathvariant="bold">n</mi><mrow><mi>i</mi><mo>+</mo><mn>1</mn></mrow></msub></mrow><mrow><mover accent="true"><mi mathvariant="bold">n</mi><mo>¯</mo></mover><mo>·</mo><msub><mi mathvariant="bold">n</mi><mrow><mi>i</mi><mo>+</mo><mn>1</mn></mrow></msub></mrow></mfrac><mo>,</mo><mover accent="true"><mi mathvariant="bold">n</mi><mo>¯</mo></mover><mo>=</mo><mfrac><mrow><msub><mi mathvariant="bold">n</mi><mi>i</mi></msub><mo>+</mo><msub><mi mathvariant="bold">n</mi><mrow><mi>i</mi><mo>+</mo><mn>1</mn></mrow></msub></mrow><mrow><mo stretchy="false">‖</mo><mrow><msub><mi mathvariant="bold">n</mi><mi>i</mi></msub><mo>+</mo><msub><mi mathvariant="bold">n</mi><mrow><mi>i</mi><mo>+</mo><mn>1</mn></mrow></msub></mrow><mo stretchy="false">‖</mo></mrow></mfrac><mo>.</mo></mrow></math></script></div><div class="l">(8)</div></div><p>
250 This is illustrated in <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F7/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F7" rid-ob="ob-F7" co-legend-rid="lgnd_F7"><span>Figure 7</span></a> (left).</p></div><!--fig ft0--><!--fig mode=article f1--><div class="fig iconblock ten_col whole_rhythm clearfix" id="F7" co-legend-rid="lgnd_F7"><a class="icnblk_img figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F7/" target="figure" rid-figpopup="F7" rid-ob="ob-F7"><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/nihms216874f7.gif" class="small-thumb" alt="Fig. 7" title="Fig. 7" src-large="/pmc/articles/PMC3179915/bin/nihms216874f7.jpg"></a><div class="icnblk_cntnt" id="lgnd_F7"><div><a class="figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F7/" target="figure" rid-figpopup="F7" rid-ob="ob-F7">Fig. 7</a></div><!--caption a4--><div><span>Left: On-the-fly interpolation between two elliptical cross-sections (<strong>ell</strong><sub><em>i</em></sub>,<strong>ell</strong><sub><em>i</em>+1</sub>), see <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#FD10" rid="FD10" class=" disp-formula">Equation 8</a>. Middle: Although this is an approximation for non-parallel (<strong>n</strong><sub><em>i</em></sub>,<strong>n</strong><sub><em>i</em>+1</sub>), the result is consistent and smooth over successive cross-sections of an axon. Gradients</span> <strong>...</strong></div></div></div><p id="P50">We compute α such that it is always 0 in the plane of <strong>ell</strong><sub><em>i</em></sub>, and 1 in the plane of <strong>ell</strong><sub><em>i</em>+1</sub>, which guarantees that successive segments between ellipse pairs line up exactly. We require a vector <strong><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/x006Ex0304.gif" border="0" alt="[n with macron]" title=""></strong> that is guaranteed not to be parallel to either ellipse, and compute α as the ratio of <em>k</em><sub>0</sub>, the distance from <strong>p</strong> along <strong><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/x006Ex0304.gif" border="0" alt="[n with macron]" title=""></strong> to <strong>ell</strong><sub><em>i</em></sub>, to <em>k</em><sub>0</sub> + <em>k</em><sub>1</sub>, the total distance between <strong>ell</strong><sub><em>i</em></sub> and <strong>ell</strong><sub><em>i</em>+1</sub> along <strong>ñ</strong> through <strong>p</strong>. We have chosen <strong>ñ</strong> as the half-way vector between <strong>n</strong><sub><em>i</em></sub> and <strong>n</strong><sub><em>i</em>+1</sub>. This choice fulfills our requirements and yields smooth results. Another obvious choice would be <strong>c</strong><sub><em>i</em>+1</sub> − <strong>c</strong><sub><em>i</em></sub>. However, in our case this vector can be close to parallel to the <strong>n</strong><sub><em>i</em></sub>, which can result in numerical problems in the denominators of <em>k</em><sub>0</sub> and <em>k</em><sub>1</sub> (<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#FD10" rid="FD10" class=" disp-formula">Equation 8</a>).</p><p id="P51">After α has been computed, it is used to obtain <strong>ell</strong>(<strong>p</strong>) as linear interpolation between the ellipse centers and axis lengths, yielding <strong>c</strong><sub><em>p</em></sub>, <span class="MathJax" id="MathJax-Element-14-Frame" role="textbox" aria-readonly="true" style=""><nobr><span class="math" id="M14" style="width: 4.393em; display: inline-block;"><span style="display: inline-block; position: relative; width: 3.58em; height: 0px; font-size: 123%;"><span style="position: absolute; clip: rect(1.846em 1000.003em 3.309em -0.431em); top: -2.762em; left: 0.003em;"><span class="mrow" id="MathJax-Span-434"><span class="mrow" id="MathJax-Span-435"><span class="msubsup" id="MathJax-Span-436"><span style="display: inline-block; position: relative; width: 0.707em; height: 0px;"><span style="position: absolute; clip: rect(1.683em 1000.003em 2.713em -0.431em); top: -2.545em; left: 0.003em;"><span class="mi" id="MathJax-Span-437" style="font-family: MathJax_Math; font-style: italic;">l</span><span style="display: inline-block; width: 0px; height: 2.55em;"></span></span><span style="position: absolute; clip: rect(1.575em 1000.003em 2.225em -0.485em); top: -2.491em; left: 0.274em;"><span class="mi" id="MathJax-Span-438" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">x</span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span><span style="position: absolute; clip: rect(1.575em 1000.003em 2.388em -0.539em); top: -1.786em; left: 0.274em;"><span class="mi" id="MathJax-Span-439" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">p</span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span></span></span><span class="mtext" id="MathJax-Span-440" style="font-family: MathJax_Main;"> and </span><span class="msubsup" id="MathJax-Span-441"><span style="display: inline-block; position: relative; width: 0.707em; height: 0px;"><span style="position: absolute; clip: rect(1.683em 1000.003em 2.713em -0.431em); top: -2.545em; left: 0.003em;"><span class="mi" id="MathJax-Span-442" style="font-family: MathJax_Math; font-style: italic;">l</span><span style="display: inline-block; width: 0px; height: 2.55em;"></span></span><span style="position: absolute; clip: rect(1.575em 1000.003em 2.388em -0.485em); top: -2.491em; left: 0.274em;"><span class="mi" id="MathJax-Span-443" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">y<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.003em;"></span></span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span><span style="position: absolute; clip: rect(1.575em 1000.003em 2.388em -0.539em); top: -1.786em; left: 0.274em;"><span class="mi" id="MathJax-Span-444" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">p</span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.767em;"></span></span></span><span style="border-left-width: 0.003em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 1.537em; vertical-align: -0.53em;"></span></span></nobr></span><script type="math/mml" id="MathJax-Element-14"><math id="M14" overflow="scroll"><mrow><msubsup><mi>l</mi><mi>p</mi><mi>x</mi></msubsup><mtext> and </mtext><msubsup><mi>l</mi><mi>p</mi><mi>y</mi></msubsup></mrow></math></script>, and spherical linear interpolation between <strong>q</strong><sub><em>i</em></sub> and <strong>q</strong><sub><sub>i</sub>+1</sub>, yielding <strong>q</strong><sub><em>p</em></sub>. Then, <strong>p</strong> is projected into the ellipse’s plane: <strong>p</strong>′ = <strong>p</strong> − <strong>n</strong><sub><em>p</em></sub> (<strong>p</strong> · <strong>n</strong><sub><em>p</em></sub> − <strong>c</strong><sub><em>p</em></sub> · <strong>n</strong><sub><em>p</em></sub>). From this, the distance value ϕ (<strong>ell</strong>(<strong>p</strong>), <strong>p</strong>′) is computed entirely in 2D in the plane of the ellipse.</p><p id="P52">This approach gives completely accurate results for parallel ellipse planes, which is a common case in axon tracking where the planes are often orthogonal to the <em>z</em> axis. It is an approximate solution for non-parallel planes that works well in practice. The angle between two successive ellipse planes <strong>n</strong><sub><em>i</em></sub> and <strong>n</strong><sub><em>i</em>+1</sub> is always quite small, even though the whole axon is allowed to curve significantly from the first cross-section to the last. <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F7/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F7" rid-ob="ob-F7" co-legend-rid="lgnd_F7"><span>Figure 7</span></a> (middle) shows a close-up of an axon with non-parallel ellipse planes, which illustrates that our approach results in visually smooth results.</p></div><div id="S22" class="sec sec-last"><p></p><h4 class="inline">5.2.2 GPU Implementation </h4><p id="P53">In order to speed up finding the two ellipses nearest to a given point <strong>p</strong> in the CUDA ray-casting kernel, ellipses are sorted into a 3D block structure (e.g., 16<sup>3</sup> blocks) before rendering that only needs to be up-dated when new ellipses are added. Each block contains links (integer indices) to all ellipses intersecting it. A single ellipse can be linked to by several blocks, but during rendering only a single block needs to be examined for each point <strong>p</strong>. In order to efficiently handle empty blocks, each block only stores the number of ellipses that intersect it and a start index into a global array of links to ellipses. The array is packed tightly such that all links of non-empty blocks are stored at consecutive memory locations. Actual ellipse information (axon-ID,<strong>c</strong><sub><em>i</em></sub>,<strong>q</strong><sub><em>i</em></sub>, <span class="MathJax" id="MathJax-Element-15-Frame" role="textbox" aria-readonly="true" style=""><nobr><span class="math" id="M15" style="width: 2.333em; display: inline-block;"><span style="display: inline-block; position: relative; width: 1.9em; height: 0px; font-size: 123%;"><span style="position: absolute; clip: rect(1.791em 1000.003em 3.255em -0.431em); top: -2.762em; left: 0.003em;"><span class="mrow" id="MathJax-Span-446"><span class="mrow" id="MathJax-Span-447"><span class="msubsup" id="MathJax-Span-448"><span style="display: inline-block; position: relative; width: 0.707em; height: 0px;"><span style="position: absolute; clip: rect(1.683em 1000.003em 2.713em -0.431em); top: -2.545em; left: 0.003em;"><span class="mi" id="MathJax-Span-449" style="font-family: MathJax_Math; font-style: italic;">l</span><span style="display: inline-block; width: 0px; height: 2.55em;"></span></span><span style="position: absolute; clip: rect(1.575em 1000.003em 2.225em -0.485em); top: -2.491em; left: 0.274em;"><span class="mi" id="MathJax-Span-450" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">x</span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span><span style="position: absolute; clip: rect(1.412em 1000.003em 2.225em -0.485em); top: -1.786em; left: 0.274em;"><span class="mi" id="MathJax-Span-451" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span></span></span><span class="mo" id="MathJax-Span-452" style="font-family: MathJax_Main;">,</span><span class="msubsup" id="MathJax-Span-453" style="padding-left: 0.165em;"><span style="display: inline-block; position: relative; width: 0.707em; height: 0px;"><span style="position: absolute; clip: rect(1.683em 1000.003em 2.713em -0.431em); top: -2.545em; left: 0.003em;"><span class="mi" id="MathJax-Span-454" style="font-family: MathJax_Math; font-style: italic;">l</span><span style="display: inline-block; width: 0px; height: 2.55em;"></span></span><span style="position: absolute; clip: rect(1.575em 1000.003em 2.388em -0.485em); top: -2.545em; left: 0.274em;"><span class="mi" id="MathJax-Span-455" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">y<span style="display: inline-block; overflow: hidden; height: 1px; width: 0.003em;"></span></span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span><span style="position: absolute; clip: rect(1.412em 1000.003em 2.225em -0.485em); top: -1.732em; left: 0.274em;"><span class="mi" id="MathJax-Span-456" style="font-size: 70.7%; font-family: MathJax_Math; font-style: italic;">i</span><span style="display: inline-block; width: 0px; height: 2.062em;"></span></span></span></span></span></span><span style="display: inline-block; width: 0px; height: 2.767em;"></span></span></span><span style="border-left-width: 0.003em; border-left-style: solid; display: inline-block; overflow: hidden; width: 0px; height: 1.537em; vertical-align: -0.463em;"></span></span></nobr></span><script type="math/mml" id="MathJax-Element-15"><math id="M15" overflow="scroll"><mrow><msubsup><mi>l</mi><mi>i</mi><mi>x</mi></msubsup><mo>,</mo><msubsup><mi>l</mi><mi>i</mi><mi>y</mi></msubsup></mrow></math></script>) is stored in a separate global ellipse array that is indexed using these links. In order to allow multiple axons to intersect the same block, multiple counts need to be stored in each block, one per axon. Furthermore, all links in a block are pre-sorted such that <strong>c</strong><sub><em>i</em>+1</sub> · <strong>n</strong><sub><em>i</em></sub> > <strong>c</strong><sub><em>i</em></sub> · <strong>n</strong><sub><em>i</em></sub>∀<sub><em>i</em></sub>, i.e., each subsequent ellipse’s center is in the front halfspace of the preceding ellipse. This simplifies the run-time search for ellipse pairs needed for interpolation, as described above. This block structure is also used for empty space skipping. Blocks with no ellipse links do not need to be searched for implicit surface intersections, and can be skipped entirely if they are transparent due to the transfer function.</p></div></div></div><div id="S23" class="sec"><div class="goto jig-ncbiinpagenav-goto-container"><a class="tgt_dark page-toc-label jig-ncbiinpagenav-goto-heading" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" title="Go to other sections in this page" role="button" aria-expanded="false" aria-haspopup="true">Go to:</a></div><h2 class="head no_bottom_margin ui-helper-clearfix" id="S23title">6 Results</h2><p id="P54">We implemented our segmentation and visualizations system on a Windows XP PC equipped with quad-core Intel Xeon 3.0 GHz CPU, 16 Gigabytes main memory, and NVIDIA Quadro 5800 and Tesla C1060 GPUs. We used a single CPU core and one GPU to compare the running time on each architecture. The CPU version is implemented using the ITK image processing library (<a href="http://www.itk.org/" ref="reftype=extlink&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CBody&TO=External%7CLink%7CURI&rendering-type=normal" target="pmc_ext">http://www.itk.org</a>). The main computational code is similar on the CPU and GPU for a fair comparison.</p><div id="S24" class="sec"><h3>6.1 Segmentation</h3><p id="P55">The running time of the CPU level set solver for 100 iterations on a 512 × 512 image is 7 seconds. It is only 0.3 second on the GPU, which shows about 23 times speed-up. Our GPU image registration runs less than a second on a 512 × 512 image (500 iterations). The total running time of our segmentation method per slice, without user interaction, is only about a second, which is sufficient for interactive applications.</p><p id="P56">To assess the performance of our segmentation method, we have segmented multiple axons in two EM datasets and measured the total and per-slice times, the amount of user intervention, and the ellipse approximation errors. The first dataset is an adult mouse cortex that consists of 101 slices of 1008 × 1065 2D image, where each pixel has five nanometers resolution and the section thickness is about 30 nanometers. The second dataset is an adult mouse hippocampus that consists of 50 slices of 1278 × 756 2D image, where each pixel is four nanometers wide and the section thickness is 29.4 nanometers. <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F8/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F8" rid-ob="ob-F8" co-legend-rid="lgnd_F8"><span>Figure 8</span></a> shows 3D renderings of the segmented axons and <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/table/T1/" target="true" class="fig-table-link table figpopup" rid-figpopup="T1" rid-ob="ob-T1" co-legend-rid=""><span>Table 1</span></a> lists the segmentation result for each dataset.</p><!--fig ft0--><!--fig mode=article f1--><div class="fig iconblock ten_col whole_rhythm clearfix" id="F8" co-legend-rid="lgnd_F8"><a class="icnblk_img figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F8/" target="figure" rid-figpopup="F8" rid-ob="ob-F8"><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/nihms216874f8.gif" class="small-thumb" alt="Fig. 8" title="Fig. 8" src-large="/pmc/articles/PMC3179915/bin/nihms216874f8.jpg"></a><div class="icnblk_cntnt" id="lgnd_F8"><div><a class="figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F8/" target="figure" rid-figpopup="F8" rid-ob="ob-F8">Fig. 8</a></div><!--caption a4--><div><span>Result images from NeuroTrace. Left: Volume rendering with edge enhancement in the upper part of the volume. Middle: Eight axons from the mouse cortex dataset. Right: Eight axons from the mouse hippocampus dataset.</span></div></div></div><!--table ft1--><!--table-wrap mode=article t1--><div class="table-wrap iconblock ten_col whole_rhythm clearfix" id="T1"><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/table/T1/" target="table" rid-ob="ob-T1" rid-figpopup="T1" class="table img_link icnblk_img figpopup"><img alt="Table 1" title="Table 1" class="small-thumb" src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/saved_resource" src-large="/pmc/articles/PMC3179915/table/T1/?report=previmg"></a><div class="icnblk_cntnt"><div><a class="figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/table/T1/" target="table" rid-figpopup="T1" rid-ob="ob-T1">Table 1</a></div><!--caption a4--><div><span>Axon segmentation results for the mouse cortex and hippocampus datasets.</span></div></div></div><p id="P57">In the mouse cortex dataset, axons A to D were traced using only axis-aligned tracking directions and axons E to H were traced using arbitrary tracking directions. All axons were traced along the z-axis in the mouse hippocampus dataset. Roughly between five to ten percent of the total number of slices were manually edited for correct segmentation for the mouse cortex dataset, and up to 20 percent of the total slices were edited on the mouse hippocampus dataset. Note that the image resolution of our input EM data is up to a factor of five higher than those used in previous work [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R13" rid="R13" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_320459230" role="button" aria-expanded="false" aria-haspopup="true">13</a>, <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R17" rid="R17" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_232989778" role="button" aria-expanded="false" aria-haspopup="true">17</a>]. The data contains more complex neural structures and is very challenging for automated methods. Total times and computing times are not significantly different between axons, and about half of the total time is used for computation.</p><p id="P58">Our ellipse-based 3D neuron representation can greatly reduce the memory footprint. For example, for an axon of 350 nm diameter we need about 70 × 70 pixels where the pixel width is 5 nm, which requires 9800 floats to store two distance fields. In contrast, to represent an equivalent 3D ellipse we only need to store nine floats, three for center and six for two axis. This yields a compression ratio of more than a factor of a thousand. <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/table/T1/" target="true" class="fig-table-link table figpopup" rid-figpopup="T1" rid-ob="ob-T1" co-legend-rid=""><span>Table 1</span></a> also shows the average distance between the ellipse and the membrane of neurons. The relative ellipse approximation errors, shown in parenthesis, range only between 0.6 to six percent of the longest axis of the ellipse, which is acceptable considering the high compression ratio we achieve.</p></div><div id="S25" class="sec"><h3>6.2 Visualization</h3><p id="P59">The prefiltering and edge-detection methods (<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F9/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F9" rid-ob="ob-F9" co-legend-rid="lgnd_F9"><span>Figure 9</span></a>) were both implemented entirely in CUDA and achieve interactive framerates. Filtering blocks on-demand and caching them for later reuse allows the user to change filters and filter settings interactively. Especially de-noising prior to calculating the edge metric improved the results considerably. The best results were achieved using anisotropic diffusion filtering. For our local histogram-based edge metric we found a histogram with 64 bins to be sufficient for our data. Also, a simple average-based histogram difference operator showed good results compared to the computationally more complex χ<sup>2</sup> distance metric. For our caching scheme we used 8<sup>3</sup> sized blocks, but this can be adjusted according to the resolution of the data. At the moment our implementation of the cache is based on CUDA arrays, but in the future we would like to use 3D textures to improve tri-linear filter performance during ray casting.</p><!--fig ft0--><!--fig mode=article f1--><div class="fig iconblock ten_col whole_rhythm clearfix" id="F9" co-legend-rid="lgnd_F9"><a class="icnblk_img figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F9/" target="figure" rid-figpopup="F9" rid-ob="ob-F9"><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/nihms216874f9.gif" class="small-thumb" alt="Fig. 9" title="Fig. 9" src-large="/pmc/articles/PMC3179915/bin/nihms216874f9.jpg"></a><div class="icnblk_cntnt" id="lgnd_F9"><div><a class="figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F9/" target="figure" rid-figpopup="F9" rid-ob="ob-F9">Fig. 9</a></div><!--caption a4--><div><span>Left: Volume Slab visualization; Top: Original data; Middle: Gradient magnitude displayed on the top slice; Bottom: Local-histogram edges; Right: Volume Rendering; Top: Original data; Middle: Gradient-magnitude shaded; Bottom: Pre-filtering and edge enhancement</span> <strong>...</strong></div></div></div><p id="P60">The dimension of EM data is highly anisotropic, with z-slice distances that can be a factor of 10 or more larger than pixel resolution. This poses real problems for volume visualization, since the visible edges from axons are shifted by large amounts between slices. Even though our filtering and edge detection method works better than traditional transfer functions, the results are sometimes still ambiguous and confusing, requiring closer inspection of the 2D slice views to identify the ROI.</p></div><div id="S26" class="sec sec-last"><h3>6.3 User Study</h3><p id="P61">We have conducted informal user studies of our segmentation method to assess the usability and accuracy of NeuroTrace by comparing it with Reconstruct [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R9" rid="R9" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_232989781" role="button" aria-expanded="false" aria-haspopup="true">9</a>]. We selected six test subjects in total. Two (Expert 1 and 2) are expert neuroscientists, and the other four (Novice 1 to 4) are novices with no previous neural process segmentation experience. We conducted two user studies, where each study required four test subjects (two experts and two novices) to perform segmentation of the same axon (axon E in the mouse cortex dataset and axon A in the mouse hippocampus dataset). We measured the total time and segmentation accuracy for both systems. We also received qualitative feedback from the users.</p><p id="P62">To measure the segmentation error, we used the Dice metric [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R8" rid="R8" class="bibr popnode jig-ncbipopper" role="button" aria-expanded="false" aria-haspopup="true">8</a>] that is commonly used to quantitatively measure the accuracy of segmentation algorithms [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R29" rid="R29" class="bibr popnode tag_hotlink tag_tooltip jig-ncbipopper" id="__tag_320459236" role="button" aria-expanded="false" aria-haspopup="true">29</a>]. The Dice metric measures similarity between two sets <em>A</em> and <em>B</em> using 2|<em>A</em> ∩ <em>B</em>|/(|<em>A</em>| + |<em>B</em>|), where | · | indicates set size. In our case, <em>A</em> is the ground truth set of pixels, and <em>B</em> is the set of pixels from the segmentation result. Dice values range between 0 and 1, where 1 implies a perfect match. We compute the Dice value for each 2D segmentation by comparing it to ground truth that was obtained by careful manual segmentation. <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/table/T2/" target="true" class="fig-table-link table figpopup" rid-figpopup="T2" rid-ob="ob-T2" co-legend-rid=""><span>Table 2</span></a> and <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/table/T3/" target="true" class="fig-table-link table figpopup" rid-figpopup="T3" rid-ob="ob-T3" co-legend-rid=""><span style="position: relative;text-decoration:none;"><span class="figpopup-sensitive-area" style="left: -1.5em;">and3</span></span><span>3</span></a> show the total segmentation times and average Dice values, and <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F10/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F10" rid-ob="ob-F10" co-legend-rid="lgnd_F10"><span>Figure 10</span></a> and <a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F11/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F11" rid-ob="ob-F11" co-legend-rid="lgnd_F11"><span style="position: relative;text-decoration:none;"><span class="figpopup-sensitive-area" style="left: -1.5em;">and11</span></span><span>11</span></a> show plots of Dice values for each slice.</p><!--fig ft0--><!--fig mode=article f1--><div class="fig iconblock ten_col whole_rhythm clearfix" id="F10" co-legend-rid="lgnd_F10"><a class="icnblk_img figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F10/" target="figure" rid-figpopup="F10" rid-ob="ob-F10"><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/nihms216874f10.gif" class="small-thumb" alt="Fig. 10" title="Fig. 10" src-large="/pmc/articles/PMC3179915/bin/nihms216874f10.jpg"></a><div class="icnblk_cntnt" id="lgnd_F10"><div><a class="figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F10/" target="figure" rid-figpopup="F10" rid-ob="ob-F10">Fig. 10</a></div><!--caption a4--><div><span>Dice value comparison of user study on the mouse cortex dataset. Left: Reconstruct. Right: NeuroTrace.</span></div></div></div><!--fig ft0--><!--fig mode=article f1--><div class="fig iconblock ten_col whole_rhythm clearfix" id="F11" co-legend-rid="lgnd_F11"><a class="icnblk_img figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F11/" target="figure" rid-figpopup="F11" rid-ob="ob-F11"><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/nihms216874f11.gif" class="small-thumb" alt="Fig. 11" title="Fig. 11" src-large="/pmc/articles/PMC3179915/bin/nihms216874f11.jpg"></a><div class="icnblk_cntnt" id="lgnd_F11"><div><a class="figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F11/" target="figure" rid-figpopup="F11" rid-ob="ob-F11">Fig. 11</a></div><!--caption a4--><div><span>Dice value comparison of user study on the mouse hippocampus dataset. Left: Reconstruct. Right: NeuroTrace.</span></div></div></div><!--table ft1--><!--table-wrap mode=article t1--><div class="table-wrap iconblock ten_col whole_rhythm clearfix" id="T2"><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/table/T2/" target="table" rid-ob="ob-T2" rid-figpopup="T2" class="table img_link icnblk_img figpopup"><img alt="Table 2" title="Table 2" class="small-thumb" src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/saved_resource(1)" src-large="/pmc/articles/PMC3179915/table/T2/?report=previmg"></a><div class="icnblk_cntnt"><div><a class="figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/table/T2/" target="table" rid-figpopup="T2" rid-ob="ob-T2">Table 2</a></div><!--caption a4--><div><span>User study results from the mouse cortex dataset.</span></div></div></div><!--table ft1--><!--table-wrap mode=article t1--><div class="table-wrap iconblock ten_col whole_rhythm clearfix" id="T3"><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/table/T3/" target="table" rid-ob="ob-T3" rid-figpopup="T3" class="table img_link icnblk_img figpopup"><img alt="Table 3" title="Table 3" class="small-thumb" src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/saved_resource(2)" src-large="/pmc/articles/PMC3179915/table/T3/?report=previmg"></a><div class="icnblk_cntnt"><div><a class="figpopup" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/table/T3/" target="table" rid-figpopup="T3" rid-ob="ob-T3">Table 3</a></div><!--caption a4--><div><span>User study results from the mouse hippocampus dataset.</span></div></div></div><p id="P63">For manual segmentation using Reconstruct there is no significant difference between the two groups in terms of the total time, but the results from the novice users are less accurate than those of the expert users. In contrast, the results using NeuroTrace do not show a significant difference between the two groups, and the novice users usually generated slightly less errors (higher Dice values) than the experts (<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/table/T2/" target="true" class="fig-table-link table figpopup" rid-figpopup="T2" rid-ob="ob-T2" co-legend-rid=""><span>Table 2</span></a>). That indicates that the semi-automated NeuroTrace is less prone to lead to human errors. In addition, NeuroTrace provides better segmentation results up to three times faster than Reconstruct. Note that Expert 2 is an exception because he spent longer time than usual and performed very accurate segmentations using Reconstruct. It is also interesting to note that the results of Reconstruct become less accurate over time, especially for novice users (<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/figure/F10/" target="figure" class="fig-table-link fig figpopup" rid-figpopup="F10" rid-ob="ob-F10" co-legend-rid="lgnd_F10"><span>Figure 10</span></a> Novice 1 and 2). This can be explained by fatigue due to the laborious manual segmentation.</p><p id="P64">The users have given highly positive feedbacks about the usability and accuracy of NeuroTrace compared to Reconstruct: “A lot easier to use; more efficient; automatic function is nice; trustworthy” (Novice 1). “Less work-demanding and accurate” (Novice 2). “Automatic segmentation was far easier to use and quicker” (Novice 3). “It is a more practical program to use and all of it’s tools are very helpful and useful” (Novice 4). “It proceeds automatically, can tilt the tracing plane” (Expert 1). “Fast, user friendly, easy to correct; visualization of the segmented data” (Expert 2). The suggestions for improvements include the addition of advanced user interface functions such as browsing of neural tracks and editing previous history, and adaptation to different data modalities, e.g., optical fluorescent confocal microcopy. Our neuroscientist collaborators are currently using NeuroTrace in their Connectomics research.</p></div></div><div id="S27" class="sec"><div class="goto jig-ncbiinpagenav-goto-container"><a class="tgt_dark page-toc-label jig-ncbiinpagenav-goto-heading" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" title="Go to other sections in this page" role="button" aria-expanded="false" aria-haspopup="true">Go to:</a></div><h2 class="head no_bottom_margin ui-helper-clearfix" id="S27title">7 Conclusions and Future Work</h2><p id="P65">In this paper we introduced NeuroTrace, a novel interactive segmentation and visualization system for neural processes in EM volumes. The main contributions are a novel semi-automatic segmentation and 3D tracking method, efficient volume rendering with on-the-fly filters and edge detection, a scalable implementation of these methods on the GPU, and a novel workflow that has been shown to be more accurate and efficient than current practice.</p><p id="P66">In the future we would like to implement a greater variety of filters and edge-detection approaches (e.g., Canny edge detection). Also we plan to automatically adjust pre-defined filter settings and opacity windowing function depending on the resolution of the input data. The biggest challenge are the extremely large z-slice distances in EM datasets. The integration of shape based-interpolation or directional coherence methods into the volume rendering might be a promising direction to solve this problem. We also would like to extend the current segmentation and tracking method to handle merging and branching of neural processes. Simultaneous tracking of multiple neural processes in a GPU cluster system would be another interesting future direction.</p></div><div id="S28" class="sec"><div class="goto jig-ncbiinpagenav-goto-container"><a class="tgt_dark page-toc-label jig-ncbiinpagenav-goto-heading" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" title="Go to other sections in this page" role="button" aria-expanded="false" aria-haspopup="true">Go to:</a></div><h2 class="head no_bottom_margin ui-helper-clearfix" id="S28title">Acknowledgments</h2><div class="sec"><p id="P67">This work was supported in part by the National Science Foundation under Grant No. PHY-0835713, the Austrian Research Promotion Agency FFG, Vienna Science and Technology Fund WWTF, the Harvard Initiative in Innovative Computing (IIC), the National Institutes of Health under Grant No. P41-RR12553-10 and U54-EB005149, and through generous support from Microsoft Research and NVIDIA. We thank our biology collaborators Prof. Jeff Lichtman and Prof. Clay Reid from the Harvard Center for Brain Science for their time and the use of their data. We also wish to thank Dr. Juan C. Tapia, Dr. Ju Lu, Thomas Zhihao Luo, May Zhang, Bo Wang, and Robert Cole Hurley for participating in the user study.</p></div></div><div id="__articleid455886aff-info" class="sec"><div class="goto jig-ncbiinpagenav-goto-container"><a class="tgt_dark page-toc-label jig-ncbiinpagenav-goto-heading" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" title="Go to other sections in this page" role="button" aria-expanded="false" aria-haspopup="true">Go to:</a></div><h2 class="head no_bottom_margin ui-helper-clearfix" id="__articleid455886aff-infotitle">Contributor Information</h2><p><span class="fm-affl">Won-Ki Jeong, </span><span class="fm-affl"> School of Engineering and Applied Sciences at Harvard University.</span></p><p><span class="fm-affl">Johanna Beyer, </span><span class="fm-affl"> VRV is Center for Virtual Reality and Visualization Research, Inc.</span></p><p><span class="fm-affl">Markus Hadwiger, </span><span class="fm-affl"> VRV is Center for Virtual Reality and Visualization Research, Inc.</span></p><p><span class="fm-affl">Amelio Vazquez, </span><span class="fm-affl"> School of Engineering and Applied Sciences at Harvard University.</span></p><p><span class="fm-affl">Hanspeter Pfister, </span><span class="fm-affl"> School of Engineering and Applied Sciences at Harvard University.</span></p><p><span class="fm-affl">Ross T. Whitaker, </span><span class="fm-affl"> Scientific Computing and Imaging Institute at the University of Utah.</span></p></div><div id="__ref-listid1073174" class="sec"><div class="goto jig-ncbiinpagenav-goto-container"><a class="tgt_dark page-toc-label jig-ncbiinpagenav-goto-heading" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" title="Go to other sections in this page" role="button" aria-expanded="false" aria-haspopup="true">Go to:</a></div><h2 class="head no_bottom_margin ui-helper-clearfix" id="__ref-listid1073174title">References</h2><div class="ref-list-sec sec" id="reference-list"><div class="ref-cit-blk half_rhythm" id="R1">1. <span>Anandan P. A computational framework and an algorithm for the measurement of visual motion. <span><span class="ref-journal">Journal on Computer Vision. </span>1989;<span class="ref-vol">2</span>:283–310.</span></span></div><div class="ref-cit-blk half_rhythm" id="R2">2. <span>Bartesaghi A, Sapiro G, Subramaniam S. An energy-based three-dimensional segmentation approach for the quantitative interpretation of electron tomograms. <span><span class="ref-journal">IEEE Trans. Image Proc. </span>2005 September;<span class="ref-vol">14</span>(9):1314–1323.</span> <span class="nowrap ref pmc">[<a class="int-reflink" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1698959/">PMC free article</a>]</span> <span class="nowrap ref pubmed">[<a href="http://www.ncbi.nlm.nih.gov/pubmed/16190467" target="pmc_ext" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CCitationRef&TO=Entrez%7CPubMed%7CRecord&rendering-type=normal">PubMed</a>]</span></span></div><div class="ref-cit-blk half_rhythm" id="R3">3. <span>Bartz D, Straßer W. Interactive exploration of extra- and intracranial blood vessels. <span><span class="ref-journal">In Proc. of IEEE Visualization. </span>1999:389–392.</span></span></div><div class="ref-cit-blk half_rhythm" id="R4">4. <span>Beyer J, Hadwiger M, Möller T, Fritz L. Smooth mixed-resolution GPU volume rendering; IEEE International Symposium on Volume and Point-Based Graphics (VG ’08); 2008. pp. 163–170.</span></div><div class="ref-cit-blk half_rhythm" id="R5">5. <span>Caban J, Rheingans P. Texture-based transfer functions for direct volume rendering. <span><span class="ref-journal">IEEE Transactions on Visualization and Computer Graphics (Proc. of IEEE Visualization ’08) </span>2008;<span class="ref-vol">14</span>(6):1364–1371.</span> <span class="nowrap ref pubmed">[<a href="http://www.ncbi.nlm.nih.gov/pubmed/18988985" target="pmc_ext" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CCitationRef&TO=Entrez%7CPubMed%7CRecord&rendering-type=normal">PubMed</a>]</span></span></div><div class="ref-cit-blk half_rhythm" id="R6">6. <span>Clarenz U, Droske M, Rumpf M. <span class="ref-journal">Inverse Problems, Image Analysis and Medical Imaging, AMS Special Session Interaction of Inverse Problems and Image Analysis.</span> volume 313. AMS; 2002. Towards fast non–rigid registration; pp. 67–84.</span></div><div class="ref-cit-blk half_rhythm" id="R7">7. <span>Deschamps T, Cohen LD. Fast extraction of minimal paths in 3d images and applications to virtual endoscopy. <span><span class="ref-journal">Medical Image Analysis. </span>2001;<span class="ref-vol">5</span>:281–299.</span> <span class="nowrap ref pubmed">[<a href="http://www.ncbi.nlm.nih.gov/pubmed/11731307" target="pmc_ext" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CCitationRef&TO=Entrez%7CPubMed%7CRecord&rendering-type=normal">PubMed</a>]</span></span></div><div class="ref-cit-blk half_rhythm" id="R8">8. <span>Dice LR. Measures of the amount of ecologic association between species. <span><span class="ref-journal">Ecology. </span>1945;<span class="ref-vol">26</span>:297–302.</span></span></div><div class="ref-cit-blk half_rhythm" id="R9">9. <span>Fiala JC. Reconstruct: a free editor for serial section microscopy. <span><span class="ref-journal">Journal of Microscopy. </span>2005 April;<span class="ref-vol">218</span>(1):52–61.</span> <span class="nowrap ref pubmed">[<a href="http://www.ncbi.nlm.nih.gov/pubmed/15817063" target="pmc_ext" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CCitationRef&TO=Entrez%7CPubMed%7CRecord&rendering-type=normal">PubMed</a>]</span></span></div><div class="ref-cit-blk half_rhythm" id="R10">10. <span>Hadwiger M, Sigg C, Scharsach H, Bühler K, Gross M. Real-time ray-casting and advanced shading of discrete isosurfaces. <span><span class="ref-journal">Computer Graphics Forum (Proc. Eurographics 2005) </span>2005;<span class="ref-vol">24</span>(3):303–312.</span></span></div><div class="ref-cit-blk half_rhythm" id="R11">11. <span>Hong L, Muraki S, Kaufman A, Bartz D, He T. Virtual voyage: interactive navigation in the human colon. In; SIGGRAPH 97 Conference Proceedings; 1997. pp. 27–34.</span></div><div class="ref-cit-blk half_rhythm" id="R12">12. <span>Jeong W-K, Whitaker RT. A fast iterative method for Eikonal equations. <span><span class="ref-journal">SIAM Journal on Scientific Computing. </span>2008;<span class="ref-vol">30</span>(5):2512–2534.</span></span></div><div class="ref-cit-blk half_rhythm" id="R13">13. <span>Jurrus E, Hardy M, Tasdizen T, Fletcher P, Koshevoy P, Chien C-B, Denk W, Whitaker R. Axon tracking in serial block-face scanning electron microscopy. <span><span class="ref-journal">Medical Image Analysis (MEDIA) </span>2009 February;<span class="ref-vol">13</span>(1):180–188.</span> <span class="nowrap ref pmc">[<a class="int-reflink" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2597704/">PMC free article</a>]</span> <span class="nowrap ref pubmed">[<a href="http://www.ncbi.nlm.nih.gov/pubmed/18617436" target="pmc_ext" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CCitationRef&TO=Entrez%7CPubMed%7CRecord&rendering-type=normal">PubMed</a>]</span></span></div><div class="ref-cit-blk half_rhythm" id="R14">14. <span>Kindlmann G, Durkin J. Semi-automatic Generation of Transfer Functions for Direct Volume Rendering. <span><span class="ref-journal">Proceedings of IEEE Volume Visualization ’98. </span>1998:79–86.</span></span></div><div class="ref-cit-blk half_rhythm" id="R15">15. <span>Kirbas C, Quek F. A review of vessel extraction techniques and algorithms. <span><span class="ref-journal">ACM Comput. Surv. </span>2004;<span class="ref-vol">36</span>(2):81–121.</span></span></div><div class="ref-cit-blk half_rhythm" id="R16">16. <span>Lefohn A, Kniss J, Hansen C, Whitaker R. Interactive deformation and visualization of level set surfaces using graphics hardware. <span><span class="ref-journal">Proceedings of IEEE Visualization. </span>2003:75–82.</span></span></div><div class="ref-cit-blk half_rhythm" id="R17">17. <span>Macke JH, Maack N, Gupta R, Denk W, Schölkopf B, Borst A. Contour-propagation algorithms for semi-automated reconstruction of neural processes. <span><span class="ref-journal">Journal of Neuroscience Methods. </span>2008;<span class="ref-vol">167</span>(2):349–357.</span> <span class="nowrap ref pubmed">[<a href="http://www.ncbi.nlm.nih.gov/pubmed/17870180" target="pmc_ext" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CCitationRef&TO=Entrez%7CPubMed%7CRecord&rendering-type=normal">PubMed</a>]</span></span></div><div class="ref-cit-blk half_rhythm" id="R18">18. <span>Maire M, Arbelaez P, Fowlkes C, Malik J. Using contours to detect and localize junctions in natural images; IEEE Conference on Computer Vision and Pattern Recognition (CVPR’08); 2008. pp. 1–8.</span></div><div class="ref-cit-blk half_rhythm" id="R19">19. <span>Martin D, Fowlkes C, Malik J. Learning to detect natural image boundaries using local brightness, color, and texture cues. <span><span class="ref-journal">IEEE Trans. on Pattern Analysis and Machine Intelligence. </span>2004;<span class="ref-vol">26</span>(1):530–549.</span> <span class="nowrap ref pubmed">[<a href="http://www.ncbi.nlm.nih.gov/pubmed/15460277" target="pmc_ext" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CCitationRef&TO=Entrez%7CPubMed%7CRecord&rendering-type=normal">PubMed</a>]</span></span></div><div class="ref-cit-blk half_rhythm" id="R20">20. <span>Mayerich D, Abbott L, Keyser J. Visualization of cellular and microvascular relationships. <span><span class="ref-journal">IEEE Transactions on Visualization and Computer Graphics. </span>2008;<span class="ref-vol">14</span>(6):1611–1618.</span> <span class="nowrap ref pubmed">[<a href="http://www.ncbi.nlm.nih.gov/pubmed/18989017" target="pmc_ext" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CCitationRef&TO=Entrez%7CPubMed%7CRecord&rendering-type=normal">PubMed</a>]</span></span></div><div class="ref-cit-blk half_rhythm" id="R21">21. <span>Mishchenko Y. Automation of 3d reconstruction of neural tissue from large volume of conventional serial section transmission electron micrographs. <span><span class="ref-journal">Journal of Neuroscience Methods. </span>2009;<span class="ref-vol">176</span>:276–289.</span> <span class="nowrap ref pmc">[<a class="int-reflink" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2948845/">PMC free article</a>]</span> <span class="nowrap ref pubmed">[<a href="http://www.ncbi.nlm.nih.gov/pubmed/18834903" target="pmc_ext" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CCitationRef&TO=Entrez%7CPubMed%7CRecord&rendering-type=normal">PubMed</a>]</span></span></div><div class="ref-cit-blk half_rhythm" id="R22">22. <span>Perona P, Malik J. Scale space and edge detection using anisotropic diffusion. <span><span class="ref-journal">IEEE Trans. in Pattern Analysis and Machine Intelligence. </span>1990;<span class="ref-vol">volume 12</span>:629–639.</span></span></div><div class="ref-cit-blk half_rhythm" id="R23">23. <span>Petrovic V, Fallon J, Kuester F. Visualizing whole-brain dti tractography with gpu-based tuboids and lod management. <span><span class="ref-journal">IEEE Trans. Vis. Comput. Graph. </span>2007;<span class="ref-vol">13</span>(6):1488–1495.</span> <span class="nowrap ref pubmed">[<a href="http://www.ncbi.nlm.nih.gov/pubmed/17968101" target="pmc_ext" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CCitationRef&TO=Entrez%7CPubMed%7CRecord&rendering-type=normal">PubMed</a>]</span></span></div><div class="ref-cit-blk half_rhythm" id="R24">24. <span>Reina G, Bidmon K, Enders F, Hastreiter P, Ertl T. GPU-Based Hyperstreamlines for Diffusion Tensor Imaging; Proceedings of EUROGRAPHICS - IEEE VGTC Symposium on Visualization 2006; 2006. pp. 35–42.</span></div><div class="ref-cit-blk half_rhythm" id="R25">25. <span>Scharsach H, Hadwiger M, Neubauer A, Bühler K. Perspective iso-surface and direct volume rendering for virtual endoscopy applications. <span><span class="ref-journal">Eurovis 2006. </span>2006:315–322.</span></span></div><div class="ref-cit-blk half_rhythm" id="R26">26. <span>Sethian J. <span class="ref-journal">Level set methods and fast marching methods.</span> Cambridge University Press; 2002. </span></div><div class="ref-cit-blk half_rhythm" id="R27">27. <span>Smith SJ. Circuit reconstruction tools today. <span><span class="ref-journal">Current Opinion in Neurobiology. </span>2007 October;<span class="ref-vol">17</span>(5):601–608.</span> <span class="nowrap ref pmc">[<a class="int-reflink" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2693015/">PMC free article</a>]</span> <span class="nowrap ref pubmed">[<a href="http://www.ncbi.nlm.nih.gov/pubmed/18082394" target="pmc_ext" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CCitationRef&TO=Entrez%7CPubMed%7CRecord&rendering-type=normal">PubMed</a>]</span></span></div><div class="ref-cit-blk half_rhythm" id="R28">28. <span>Sporns O, Tononi G, Kötter R. The human connectome: A structural description of the human brain. <span><span class="ref-journal">PLoS Computational Biology. </span>2005 September;<span class="ref-vol">1</span>(4):e42+.</span> <span class="nowrap ref pmc">[<a class="int-reflink" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1239902/">PMC free article</a>]</span> <span class="nowrap ref pubmed">[<a href="http://www.ncbi.nlm.nih.gov/pubmed/16201007" target="pmc_ext" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CCitationRef&TO=Entrez%7CPubMed%7CRecord&rendering-type=normal">PubMed</a>]</span></span></div><div class="ref-cit-blk half_rhythm" id="R29">29. <span>Tasdizen T, Awate S, Whitaker R, Foster N. MRI tissue classification with neighborhood statistics: A nonparametric, entropy-minimizing approach. <span><span class="ref-journal">MICCAI 2005. </span>2005:517–525.</span> <span class="nowrap ref pubmed">[<a href="http://www.ncbi.nlm.nih.gov/pubmed/16685999" target="pmc_ext" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article%7CCitationRef&TO=Entrez%7CPubMed%7CRecord&rendering-type=normal">PubMed</a>]</span></span></div><div class="ref-cit-blk half_rhythm" id="R30">30. <span>Tasdizen T, Whitaker R, Marc R, Jones B. Enhancement of cell boundaries in transmission micropscopy images; IEEE International Conf. on Image Processing (ICIP ’05); 2005. pp. 129–132.</span></div><div class="ref-cit-blk half_rhythm" id="R31">31. <span>Tomasi C, Manduchi R. Bilateral filtering for gray and color images. <span><span class="ref-journal">ICCV ’98. </span>1998:839–846.</span></span></div><div class="ref-cit-blk half_rhythm" id="R32">32. <span>Vazquez-Reina A, Miller E, Pfister H. Multiphase geometric couplings for the segmentation of neural processes; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR); 2009. pp. 2020–2027.</span></div></div></div><div style="display: none; width: 200px; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true" class="ui-helper-reset ui-ncbipopper-wrapper ui-ncbilinksmenu"><ul id="ui-ncbiinpagenav-1" style=""><li><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#__abstractid735706title">Abstract</a></li><li><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#S1title">1 Introduction</a></li><li><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#S2title">2 Previous Work</a></li><li><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#S7title">3 Workflow</a></li><li><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#S10title">4 Segmentation</a></li><li><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#S14title">5 Volume Visualization</a></li><li><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#S23title">6 Results</a></li><li><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#S27title">7 Conclusions and Future Work</a></li><li><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#S28title">Acknowledgments</a></li><li><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#__articleid455886aff-infotitle">Contributor Information</a></li><li><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#__ref-listid1073174title">References</a></li></ul></div></div><!--post-content--></div>
256 <!-- Book content -->
259 <div id="rightcolumn" class="four_col col last">
260 <!-- Custom content above discovery portlets -->
265 <div><div class="try-button"><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/?report=reader"><img src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/3838809"></a></div><div class="format-menu"><h2>Formats:</h2><ul><li class="selected">Article</li> | <li><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/?report=reader">PubReader</a></li> | <li class="epub-link"><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/epub/">ePub (beta)</a></li> | <li><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/pdf/nihms216874.pdf">PDF (5.2M)</a></li></ul></div></div><div class="portlet"><div class="portlet_head"><div class="portlet_title"><h3><span>Related citations in PubMed</span></h3></div><a name="Shutter" sid="1" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" class="portlet_shutter ui-ncbitoggler-open" title="Show/hide content" remembercollapsed="true" pgsec_name="PBooksDiscovery_RA" id="Shutter" toggles="ui-portlet_content-2" aria-disabled="false" role="button" aria-expanded="true" style="position: absolute; padding: 0px;"></a></div><div class="ui-helper-reset" aria-live="assertive"><div class="portlet_content ui-ncbitoggler ui-ncbitoggler-slave-open" id="ui-portlet_content-2" aria-hidden="false"><ul><li class="brieflinkpopper two_line"><a class="brieflinkpopperctrl" href="http://www.ncbi.nlm.nih.gov/pubmed/18243401" ref="reftype=relart&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|RelatedArticles&TO=Entrez|Pubmed|Record&rendering-type=normal" role="button" aria-expanded="false" aria-haspopup="true">Real-time visualization of large volume datasets on standard PC hardware.</a><span class="source">[Comput Methods Programs Biomed...]</span><div class="brieflinkpop ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" id="ui-brieflinkpop-16" aria-live="assertive" aria-hidden="true" style="top: -100px; left: -100px; display: none;"><em class="author">Xie KYang J, Zhu YM, . </em><em class="cit">Comput Methods Programs Biomed. 2008 May; 90(2):117-23. Epub 2008 Feb 19.</em></div></li><li class="brieflinkpopper two_line"><a class="brieflinkpopperctrl" href="http://www.ncbi.nlm.nih.gov/pubmed/20650718" ref="reftype=relart&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|RelatedArticles&TO=Entrez|Pubmed|Record&rendering-type=normal" role="button" aria-expanded="false" aria-haspopup="true">Ssecrett and NeuroTrace: interactive visualization and analysis tools for large-scale neuroscience data sets.</a><span class="source">[IEEE Comput Graph Appl. 2010]</span><div class="brieflinkpop ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" id="ui-brieflinkpop-17" aria-live="assertive" aria-hidden="true" style="top: -100px; left: -100px; display: none;"><em class="author">Jeong WKBeyer J, Hadwiger M, Blue R, Law C, Vazquez-Reina A, Reid RC, Lichtman J, Pfister H, . </em><em class="cit">IEEE Comput Graph Appl. 2010 May-Jun; 30(3):58-70. </em></div></li><li class="brieflinkpopper two_line"><a class="brieflinkpopperctrl" href="http://www.ncbi.nlm.nih.gov/pubmed/22003674" ref="reftype=relart&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|RelatedArticles&TO=Entrez|Pubmed|Record&rendering-type=normal" role="button" aria-expanded="false" aria-haspopup="true">Carving: scalable interactive segmentation of neural volume electron microscopy images.</a><span class="source">[Med Image Comput Comput Assist...]</span><div class="brieflinkpop ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" id="ui-brieflinkpop-18" aria-live="assertive" aria-hidden="true" style="top: -100px; left: -100px; display: none;"><em class="author">Straehle CNKöthe U, Knott G, Hamprecht FA, . </em><em class="cit">Med Image Comput Comput Assist Interv. 2011; 14(Pt 1):653-60. </em></div></li><li class="brieflinkpopper two_line"><a class="brieflinkpopperctrl" href="http://www.ncbi.nlm.nih.gov/pubmed/22119321" ref="reftype=relart&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|RelatedArticles&TO=Entrez|Pubmed|Record&rendering-type=normal" role="button" aria-expanded="false" aria-haspopup="true">Volume electron microscopy for neuronal circuit reconstruction.</a><span class="source">[Curr Opin Neurobiol. 2012]</span><div class="brieflinkpop ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" id="ui-brieflinkpop-19" aria-live="assertive" aria-hidden="true" style="top: -100px; left: -100px; display: none;"><em class="author">Briggman KLBock DD, . </em><em class="cit">Curr Opin Neurobiol. 2012 Feb; 22(1):154-61. Epub 2011 Nov 24.</em></div></li><li class="brieflinkpopper two_line"><a class="brieflinkpopperctrl" href="http://www.ncbi.nlm.nih.gov/pubmed/20714917" ref="reftype=relart&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|RelatedArticles&TO=Entrez|Pubmed|Record&rendering-type=normal" role="button" aria-expanded="false" aria-haspopup="true">Volume visualization: a technical overview with a focus on medical applications.</a><span class="source">[J Digit Imaging. 2011]</span><div class="brieflinkpop ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" id="ui-brieflinkpop-20" aria-live="assertive" aria-hidden="true" style="top: -100px; left: -100px; display: none;"><em class="author">Zhang QEagleson R, Peters TM, . </em><em class="cit">J Digit Imaging. 2011 Aug; 24(4):640-64. </em></div></li></ul><a class="seemore" href="http://www.ncbi.nlm.nih.gov/sites/entrez?db=pubmed&cmd=link&linkname=pubmed_pubmed_reviews&uid=19834227&log$=relatedreviews&logdbfrom=pmc" ref="reftype=relart&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|RelatedArticles&TO=Entrez|Pubmed|Reviews&rendering-type=normal">See reviews...</a><a class="seemore" href="http://www.ncbi.nlm.nih.gov/sites/entrez?db=pubmed&cmd=link&linkname=pubmed_pubmed&uid=19834227&log$=relatedarticles&logdbfrom=pmc" ref="reftype=relart&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|RelatedArticles&TO=Entrez|Pubmed|Related%20Records&rendering-type=normal">See all...</a></div></div></div><div class="portlet"><div class="portlet_head"><div class="portlet_title"><h3><span>Cited by other articles in PMC</span></h3></div><a name="Shutter" sid="1" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" class="portlet_shutter ui-ncbitoggler-open" title="Show/hide content" remembercollapsed="true" pgsec_name="PMCVCitedByPmcArticlesP" id="Shutter" toggles="ui-portlet_content-3" aria-disabled="false" role="button" aria-expanded="true" style="position: absolute; padding: 0px;"></a></div><div class="ui-helper-reset" aria-live="assertive"><div class="portlet_content ui-ncbitoggler ui-ncbitoggler-slave-open" id="ui-portlet_content-3" aria-hidden="false"><ul><li class="brieflinkpopper two_line"><a class="brieflinkpopperctrl" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3681936/" ref="reftype=relart&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedByInPmc&TO=PMC|Article|Record&rendering-type=normal" role="button" aria-expanded="false" aria-haspopup="true">Computational Tissue Volume Reconstruction of a Peripheral Nerve Using High-Resolution Light-Microscopy and Reconstruct</a><span class="source">[PLoS ONE. ]</span><div class="brieflinkpop ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" id="ui-brieflinkpop-21" aria-live="assertive" aria-hidden="true" style="top: -100px; left: -100px; display: none;"><em class="author">Gierthmuehlen M, Freiman TM, Haastert-Talini K, Mueller A, Kaminsky J, Stieglitz T, Plachta DT. </em><em class="cit">PLoS ONE. 8(6)e66191</em></div></li><li class="brieflinkpopper two_line"><a class="brieflinkpopperctrl" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3654713/" ref="reftype=relart&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedByInPmc&TO=PMC|Article|Record&rendering-type=normal" role="button" aria-expanded="false" aria-haspopup="true">DP2: Distributed 3D image segmentation using micro-labor workforce</a><span class="source">[Bioinformatics. 2013]</span><div class="brieflinkpop ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" id="ui-brieflinkpop-22" aria-live="assertive" aria-hidden="true" style="top: -100px; left: -100px; display: none;"><em class="author">Giuly RJ, Kim KY, Ellisman MH. </em><em class="cit">Bioinformatics. 2013 May 15; 29(10)1359-1360</em></div></li><li class="brieflinkpopper two_line"><a class="brieflinkpopperctrl" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3546938/" ref="reftype=relart&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedByInPmc&TO=PMC|Article|Record&rendering-type=normal" role="button" aria-expanded="false" aria-haspopup="true">Computer Assisted Assembly of Connectomes from Electron Micrographs: Application to Caenorhabditis elegans</a><span class="source">[PLoS ONE. ]</span><div class="brieflinkpop ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" id="ui-brieflinkpop-23" aria-live="assertive" aria-hidden="true" style="top: -100px; left: -100px; display: none;"><em class="author">Xu M, Jarrell TA, Wang Y, Cook SJ, Hall DH, Emmons SW. </em><em class="cit">PLoS ONE. 8(1)e54050</em></div></li><li class="brieflinkpopper two_line"><a class="brieflinkpopperctrl" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3496509/" ref="reftype=relart&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedByInPmc&TO=PMC|Article|Record&rendering-type=normal" role="button" aria-expanded="false" aria-haspopup="true">A survey of GPU-based medical image computing techniques</a><span class="source">[Quantitative Imaging in Medici...]</span><div class="brieflinkpop ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" id="ui-brieflinkpop-24" aria-live="assertive" aria-hidden="true" style="top: -100px; left: -100px; display: none;"><em class="author">Shi L, Liu W, Zhang H, Xie Y, Wang D. </em><em class="cit">Quantitative Imaging in Medicine and Surgery. 2012 Sep; 2(3)188-206</em></div></li><li class="brieflinkpopper two_line"><a class="brieflinkpopperctrl" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2975605/" ref="reftype=relart&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedByInPmc&TO=PMC|Article|Record&rendering-type=normal" role="button" aria-expanded="false" aria-haspopup="true">Machines that learn to segment images: a crucial technology for connectomics</a><span class="source">[Current opinion in neurobiolog...]</span><div class="brieflinkpop ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" id="ui-brieflinkpop-25" aria-live="assertive" aria-hidden="true" style="top: -100px; left: -100px; display: none;"><em class="author">Jain V, Seung HS, Turaga SC. </em><em class="cit">Current opinion in neurobiology. 2010 Oct; 20(5)653-666</em></div></li></ul><a class="seemore" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/citedby/" ref="reftype=relart&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedByInPmc&TO=PMC|Article|CitedBy&rendering-type=normal">See all...</a></div></div></div><div class="portlet"><div class="portlet_head"><div class="portlet_title"><h3><span>Links</span></h3></div><a name="Shutter" sid="1" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" class="portlet_shutter ui-ncbitoggler-open" title="Show/hide content" remembercollapsed="true" pgsec_name="PMCVCitedByPmcArticlesP" id="Shutter" toggles="ui-portlet_content-4" aria-disabled="false" role="button" aria-expanded="true" style="position: absolute; padding: 0px;"></a></div><div class="ui-helper-reset" aria-live="assertive"><div class="portlet_content ui-ncbitoggler ui-ncbitoggler-slave-open" id="ui-portlet_content-4" aria-hidden="false"><ul><li class="brieflinkpopper"><a class="brieflinkpopperctrl" href="http://www.ncbi.nlm.nih.gov/pubmed/19834227/" ref="reftype=PubMed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|DiscoveryLinks&TO=Entrez|PubMed|Record&rendering-type=normal" role="button" aria-expanded="false" aria-haspopup="true">PubMed</a><div class="brieflinkpop ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" id="ui-brieflinkpop-26" aria-live="assertive" aria-hidden="true" style="top: -100px; left: -100px; display: none;">PubMed<div class="brieflinkpopdesc">PubMed citations for these articles</div></div></li><li class="brieflinkpopper"><a class="brieflinkpopperctrl" href="http://www.ncbi.nlm.nih.gov/pmc/?Db=taxonomy&DbFrom=pmc&Cmd=Link&LinkName=pmc_taxonomy&IdsFromResult=3179915" ref="reftype=Taxonomy&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|DiscoveryLinks&TO=Entrez|Crosslink|Taxonomy&rendering-type=normal" role="button" aria-expanded="false" aria-haspopup="true">Taxonomy</a><div class="brieflinkpop ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" id="ui-brieflinkpop-27" aria-live="assertive" aria-hidden="true" style="top: -100px; left: -100px; display: none;">Taxonomy<div class="brieflinkpopdesc">Related taxonomy entry</div></div></li><li class="brieflinkpopper"><a class="brieflinkpopperctrl" href="http://www.ncbi.nlm.nih.gov/pmc/?Db=taxonomy&DbFrom=pmc&cmd=Link&LinkName=pmc_taxonomy&doptcmdl=TxTree&IdsFromResult=3179915" ref="reftype=Taxonomy%20Tree&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|DiscoveryLinks&TO=Entrez|Crosslink|Taxonomy&rendering-type=normal" role="button" aria-expanded="false" aria-haspopup="true">Taxonomy Tree</a><div class="brieflinkpop ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" id="ui-brieflinkpop-28" aria-live="assertive" aria-hidden="true" style="top: -100px; left: -100px; display: none;">Taxonomy Tree</div></li></ul></div></div></div><div class="portlet"><div class="portlet_head"><div class="portlet_title"><h3><span>Recent activity</span></h3></div><a name="Shutter" sid="1" href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" class="portlet_shutter ui-ncbitoggler-open" title="Show/hide content" remembercollapsed="true" pgsec_name="recent_activity" id="Shutter" toggles="ui-portlet_content-5" aria-disabled="false" role="button" aria-expanded="true" style="position: absolute; padding: 0px;"></a></div><div class="ui-helper-reset" aria-live="assertive"><div class="portlet_content ui-ncbitoggler ui-ncbitoggler-slave-open" id="ui-portlet_content-5" aria-hidden="false"><div id="HTDisplay" class=""><div class="action"><a href="javascript:historyDisplayState('ClearHT')">Clear</a><a href="javascript:historyDisplayState('HTOff')" class="HTOn">Turn Off</a><a href="javascript:historyDisplayState('HTOn')" class="HTOff">Turn On</a></div><ul id="activity"><li class="ra_rcd ralinkpopper two_line" role="button" aria-expanded="false" aria-haspopup="true"><a class="htb ralinkpopperctrl" ref="log$=activity&linkpos=1" href="http://www.ncbi.nlm.nih.gov/portal/utils/pageresolver.fcgi?recordid=1379421854219750">Scalable and Interactive Segmentation and Visualization of Neural Processes in E...</a><div class="ralinkpop ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" id="ui-ralinkpop-29" aria-live="assertive" aria-hidden="true" style="top: -100px; left: -100px; display: none;">Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets<div class="brieflinkpopdesc">NIHPA Author Manuscripts. Nov-Dec 2009; 15(6)1505</div></div><div class="tertiary">PMC</div></li></ul><p class="HTOn">Your browsing activity is empty.</p><p class="HTOff">Activity recording is turned off.</p><p id="turnOn" class="HTOff"><a href="javascript:historyDisplayState('HTOn')">Turn recording back on</a></p><a class="seemore" href="http://www.ncbi.nlm.nih.gov/sites/myncbi/recentactivity">See more...</a></div></div></div></div>
267 <div><div class="portlet brieflink pmc_para_cit" id="crb--P2" name="crb--P2" rid="P2" style="position: absolute; visibility: visible; width: 370px; top: 976px; display: none;"><div class="portlet_head"></div><div class="portlet_content" id="ui-portlet_content-6"><ul><li class="two_line" reference_id="R28"><a href="http://www.ncbi.nlm.nih.gov/pubmed/16201007/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal"><span class="invert">Review</span> The human connectome: A structural description of the human brain.</a><span class="one_line_source">[PLoS Comput Biol. 2005]</span><div class="alt-note"><div class="authors">Sporns O, Tononi G, Kötter R</div><div class="citation">PLoS Comput Biol. 2005 Sep; 1(4):e42.</div></div></li></ul></div></div><div class="portlet brieflink pmc_para_cit" id="crb--P5" name="crb--P5" rid="P5" style="position: absolute; visibility: visible; width: 370px; top: 1986px;"><div class="portlet_head"></div><div class="portlet_content" id="ui-portlet_content-7"><ul><li class="two_line" reference_id="R13"><a href="http://www.ncbi.nlm.nih.gov/pubmed/18617436/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal">Axon tracking in serial block-face scanning electron microscopy.</a><span class="one_line_source">[Med Image Anal. 2009]</span><div class="alt-note"><div class="authors">Jurrus E, Hardy M, Tasdizen T, Fletcher PT, Koshevoy P, Chien CB, Denk W, Whitaker R</div><div class="citation">Med Image Anal. 2009 Feb; 13(1):180-8.</div></div></li><li class="two_line" reference_id="R17"><a href="http://www.ncbi.nlm.nih.gov/pubmed/17870180/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal">Contour-propagation algorithms for semi-automated reconstruction of neural processes.</a><span class="one_line_source">[J Neurosci Methods. 2008]</span><div class="alt-note"><div class="authors">Macke JH, Maack N, Gupta R, Denk W, Schölkopf B, Borst A</div><div class="citation">J Neurosci Methods. 2008 Jan 30; 167(2):349-57.</div></div></li><li class="two_line" reference_id="R21"><a href="http://www.ncbi.nlm.nih.gov/pubmed/18834903/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal">Automation of 3D reconstruction of neural tissue from large volume of conventional serial section transmission electron micrographs.</a><span class="one_line_source">[J Neurosci Methods. 2009]</span><div class="alt-note"><div class="authors">Mishchenko Y</div><div class="citation">J Neurosci Methods. 2009 Jan 30; 176(2):276-89.</div></div></li><li class="two_line" reference_id="R2"><a href="http://www.ncbi.nlm.nih.gov/pubmed/16190467/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal">An energy-based three-dimensional segmentation approach for the quantitative interpretation of electron tomograms.</a><span class="one_line_source">[IEEE Trans Image Process. 2005]</span><div class="alt-note"><div class="authors">Bartesaghi A, Sapiro G, Subramaniam S</div><div class="citation">IEEE Trans Image Process. 2005 Sep; 14(9):1314-23.</div></div></li></ul></div></div><div class="portlet brieflink pmc_para_cit" id="crb--P6" name="crb--P6" rid="P6" style="position: absolute; visibility: visible; width: 370px; top: 2325px;"><div class="portlet_head"></div><div class="portlet_content" id="ui-portlet_content-8"><ul><li class="two_line" reference_id="R7"><a href="http://www.ncbi.nlm.nih.gov/pubmed/11731307/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal">Fast extraction of minimal paths in 3D images and applications to virtual endoscopy.</a><span class="one_line_source">[Med Image Anal. 2001]</span><div class="alt-note"><div class="authors">Deschamps T, Cohen LD</div><div class="citation">Med Image Anal. 2001 Dec; 5(4):281-99.</div></div></li></ul></div></div><div class="portlet brieflink pmc_para_cit" id="crb--P7" name="crb--P7" rid="P7" style="position: absolute; visibility: visible; width: 370px; top: 2475px;"><div class="portlet_head"></div><div class="portlet_content" id="ui-portlet_content-9"><ul><li class="two_line" reference_id="R20"><a href="http://www.ncbi.nlm.nih.gov/pubmed/18989017/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal">Visualization of cellular and microvascular relationships.</a><span class="one_line_source">[IEEE Trans Vis Comput Graph. 2008]</span><div class="alt-note"><div class="authors">Mayerich DM, Abbott L, Keyser J</div><div class="citation">IEEE Trans Vis Comput Graph. 2008 Nov-Dec; 14(6):1611-8.</div></div></li><li class="two_line" reference_id="R5"><a href="http://www.ncbi.nlm.nih.gov/pubmed/18988985/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal">Texture-based transfer functions for direct volume rendering.</a><span class="one_line_source">[IEEE Trans Vis Comput Graph. 2008]</span><div class="alt-note"><div class="authors">Caban JJ, Rheingans P</div><div class="citation">IEEE Trans Vis Comput Graph. 2008 Nov-Dec; 14(6):1364-71.</div></div></li><li class="two_line" reference_id="R19"><a href="http://www.ncbi.nlm.nih.gov/pubmed/15460277/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal">Learning to detect natural image boundaries using local brightness, color, and texture cues.</a><span class="one_line_source">[IEEE Trans Pattern Anal Mach Intell. 2004]</span><div class="alt-note"><div class="authors">Martin DR, Fowlkes CC, Malik J</div><div class="citation">IEEE Trans Pattern Anal Mach Intell. 2004 May; 26(5):530-49.</div></div></li></ul></div></div><div class="portlet brieflink pmc_para_cit" id="crb--P8" name="crb--P8" rid="P8" style="position: absolute; visibility: visible; width: 370px; top: 2752px;"><div class="portlet_head"></div><div class="portlet_content" id="ui-portlet_content-10"><ul><li class="two_line" reference_id="R23"><a href="http://www.ncbi.nlm.nih.gov/pubmed/17968101/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal">Visualizing whole-brain DTI tractography with GPU-based Tuboids and LoD management.</a><span class="one_line_source">[IEEE Trans Vis Comput Graph. 2007]</span><div class="alt-note"><div class="authors">Petrovic V, Fallon J, Kuester F</div><div class="citation">IEEE Trans Vis Comput Graph. 2007 Nov-Dec; 13(6):1488-95.</div></div></li></ul></div></div><div class="portlet brieflink pmc_para_cit" id="crb--P10" name="crb--P10" rid="P10" style="position: absolute; visibility: visible; width: 370px; top: 3034px;"><div class="portlet_head"></div><div class="portlet_content" id="ui-portlet_content-11"><ul><li class="two_line" reference_id="R27"><a href="http://www.ncbi.nlm.nih.gov/pubmed/18082394/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal"><span class="invert">Review</span> Circuit reconstruction tools today.</a><span class="one_line_source">[Curr Opin Neurobiol. 2007]</span><div class="alt-note"><div class="authors">Smith SJ</div><div class="citation">Curr Opin Neurobiol. 2007 Oct; 17(5):601-8.</div></div></li><li class="two_line" reference_id="R9"><a href="http://www.ncbi.nlm.nih.gov/pubmed/15817063/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal">Reconstruct: a free editor for serial section microscopy.</a><span class="one_line_source">[J Microsc. 2005]</span><div class="alt-note"><div class="authors">Fiala JC</div><div class="citation">J Microsc. 2005 Apr; 218(Pt 1):52-61.</div></div></li></ul></div></div><div class="portlet brieflink pmc_para_cit" id="crb--P36" name="crb--P36" rid="P36" style="position: absolute; visibility: visible; width: 370px; top: 8311px;"><div class="portlet_head"></div><div class="portlet_content" id="ui-portlet_content-12"><ul><li class="two_line" reference_id="R19"><a href="http://www.ncbi.nlm.nih.gov/pubmed/15460277/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal">Learning to detect natural image boundaries using local brightness, color, and texture cues.</a><span class="one_line_source">[IEEE Trans Pattern Anal Mach Intell. 2004]</span><div class="alt-note"><div class="authors">Martin DR, Fowlkes CC, Malik J</div><div class="citation">IEEE Trans Pattern Anal Mach Intell. 2004 May; 26(5):530-49.</div></div></li></ul></div></div><div class="portlet brieflink pmc_para_cit" id="crb--P57" name="crb--P57" rid="P57" style="position: absolute; visibility: visible; width: 370px; top: 12766px;"><div class="portlet_head"></div><div class="portlet_content" id="ui-portlet_content-13"><ul><li class="two_line" reference_id="R13"><a href="http://www.ncbi.nlm.nih.gov/pubmed/18617436/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal">Axon tracking in serial block-face scanning electron microscopy.</a><span class="one_line_source">[Med Image Anal. 2009]</span><div class="alt-note"><div class="authors">Jurrus E, Hardy M, Tasdizen T, Fletcher PT, Koshevoy P, Chien CB, Denk W, Whitaker R</div><div class="citation">Med Image Anal. 2009 Feb; 13(1):180-8.</div></div></li><li class="two_line" reference_id="R17"><a href="http://www.ncbi.nlm.nih.gov/pubmed/17870180/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal">Contour-propagation algorithms for semi-automated reconstruction of neural processes.</a><span class="one_line_source">[J Neurosci Methods. 2008]</span><div class="alt-note"><div class="authors">Macke JH, Maack N, Gupta R, Denk W, Schölkopf B, Borst A</div><div class="citation">J Neurosci Methods. 2008 Jan 30; 167(2):349-57.</div></div></li></ul></div></div><div class="portlet brieflink pmc_para_cit" id="crb--P61" name="crb--P61" rid="P61" style="position: absolute; visibility: visible; width: 370px; top: 13707px;"><div class="portlet_head"></div><div class="portlet_content" id="ui-portlet_content-14"><ul><li class="two_line" reference_id="R9"><a href="http://www.ncbi.nlm.nih.gov/pubmed/15817063/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal">Reconstruct: a free editor for serial section microscopy.</a><span class="one_line_source">[J Microsc. 2005]</span><div class="alt-note"><div class="authors">Fiala JC</div><div class="citation">J Microsc. 2005 Apr; 218(Pt 1):52-61.</div></div></li></ul></div></div><div class="portlet brieflink pmc_para_cit" id="crb--P62" name="crb--P62" rid="P62" style="position: absolute; visibility: visible; width: 370px; top: 13885px;"><div class="portlet_head"></div><div class="portlet_content" id="ui-portlet_content-15"><ul><li class="two_line" reference_id="R29"><a href="http://www.ncbi.nlm.nih.gov/pubmed/16685999/" ref="reftype=pubmed&article-id=3179915&issue-id=200352&journal-id=319&FROM=Article|CitedRefBlock&TO=Entrez|Pubmed|Record&rendering-type=normal">MRI tissue classification with neighborhood statistics: a nonparametric, entropy-minimizing approach.</a><span class="one_line_source">[Med Image Comput Comput Assist Interv. 2005]</span><div class="alt-note"><div class="authors">Tasdizen T, Awate SP, Whitaker RT, Foster NL</div><div class="citation">Med Image Comput Comput Assist Interv. 2005; 8(Pt 2):517-25.</div></div></li></ul></div></div></div>
269 <!-- Custom content below discovery portlets -->
276 <!-- Custom content after all -->
284 <script src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/jquery.scrollTo-1.4.2.js"></script>
287 $('.skiplink').each(function(i, item){
288 var href = $($(item).attr('href'));
289 href.attr('tabindex', '-1').addClass('skiptarget'); // ensure the target can receive focus
290 $(item).on('click', function(event){
291 event.preventDefault();
292 $.scrollTo(href, 0, {
304 <div id="body-link-poppers"><div id="body-link-popper-R28" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true"><span class="invert">Review</span> The human connectome: A structural description of the human brain.<div class="authors">Sporns O, Tononi G, Kötter R</div><div class="citation">PLoS Comput Biol. 2005 Sep; 1(4):e42.</div><p>[<a href="http://www.ncbi.nlm.nih.gov/pubmed/16201007/">PubMed</a>] [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R28">Ref list</a>]</p></div><div id="body-link-popper-R32" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">32. Vazquez-Reina A, Miller E, Pfister H. Multiphase geometric couplings for the segmentation of neural processes; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR); 2009. pp. 2020–2027. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R32">Ref list</a>]</div><div id="body-link-popper-R13" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">Axon tracking in serial block-face scanning electron microscopy.<div class="authors">Jurrus E, Hardy M, Tasdizen T, Fletcher PT, Koshevoy P, Chien CB, Denk W, Whitaker R</div><div class="citation">Med Image Anal. 2009 Feb; 13(1):180-8.</div><p>[<a href="http://www.ncbi.nlm.nih.gov/pubmed/18617436/">PubMed</a>] [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R13">Ref list</a>]</p></div><div id="body-link-popper-R17" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">Contour-propagation algorithms for semi-automated reconstruction of neural processes.<div class="authors">Macke JH, Maack N, Gupta R, Denk W, Schölkopf B, Borst A</div><div class="citation">J Neurosci Methods. 2008 Jan 30; 167(2):349-57.</div><p>[<a href="http://www.ncbi.nlm.nih.gov/pubmed/17870180/">PubMed</a>] [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R17">Ref list</a>]</p></div><div id="body-link-popper-R21" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">Automation of 3D reconstruction of neural tissue from large volume of conventional serial section transmission electron micrographs.<div class="authors">Mishchenko Y</div><div class="citation">J Neurosci Methods. 2009 Jan 30; 176(2):276-89.</div><p>[<a href="http://www.ncbi.nlm.nih.gov/pubmed/18834903/">PubMed</a>] [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R21">Ref list</a>]</p></div><div id="body-link-popper-R2" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">An energy-based three-dimensional segmentation approach for the quantitative interpretation of electron tomograms.<div class="authors">Bartesaghi A, Sapiro G, Subramaniam S</div><div class="citation">IEEE Trans Image Process. 2005 Sep; 14(9):1314-23.</div><p>[<a href="http://www.ncbi.nlm.nih.gov/pubmed/16190467/">PubMed</a>] [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R2">Ref list</a>]</p></div><div id="body-link-popper-R15" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">15. Kirbas C, Quek F. A review of vessel extraction techniques and algorithms. ACM Comput. Surv. 2004;36(2):81–121. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R15">Ref list</a>]</div><div id="body-link-popper-R11" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">11. Hong L, Muraki S, Kaufman A, Bartz D, He T. Virtual voyage: interactive navigation in the human colon. In; SIGGRAPH 97 Conference Proceedings; 1997. pp. 27–34. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R11">Ref list</a>]</div><div id="body-link-popper-R3" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">3. Bartz D, Straßer W. Interactive exploration of extra- and intracranial blood vessels. In Proc. of IEEE Visualization. 1999:389–392. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R3">Ref list</a>]</div><div id="body-link-popper-R7" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">Fast extraction of minimal paths in 3D images and applications to virtual endoscopy.<div class="authors">Deschamps T, Cohen LD</div><div class="citation">Med Image Anal. 2001 Dec; 5(4):281-99.</div><p>[<a href="http://www.ncbi.nlm.nih.gov/pubmed/11731307/">PubMed</a>] [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R7">Ref list</a>]</p></div><div id="body-link-popper-R20" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">Visualization of cellular and microvascular relationships.<div class="authors">Mayerich DM, Abbott L, Keyser J</div><div class="citation">IEEE Trans Vis Comput Graph. 2008 Nov-Dec; 14(6):1611-8.</div><p>[<a href="http://www.ncbi.nlm.nih.gov/pubmed/18989017/">PubMed</a>] [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R20">Ref list</a>]</p></div><div id="body-link-popper-R25" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">25. Scharsach H, Hadwiger M, Neubauer A, Bühler K. Perspective iso-surface and direct volume rendering for virtual endoscopy applications. Eurovis 2006. 2006:315–322. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R25">Ref list</a>]</div><div id="body-link-popper-R4" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">4. Beyer J, Hadwiger M, Möller T, Fritz L. Smooth mixed-resolution GPU volume rendering; IEEE International Symposium on Volume and Point-Based Graphics (VG ’08); 2008. pp. 163–170. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R4">Ref list</a>]</div><div id="body-link-popper-R14" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">14. Kindlmann G, Durkin J. Semi-automatic Generation of Transfer Functions for Direct Volume Rendering. Proceedings of IEEE Volume Visualization ’98. 1998:79–86. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R14">Ref list</a>]</div><div id="body-link-popper-R5" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">Texture-based transfer functions for direct volume rendering.<div class="authors">Caban JJ, Rheingans P</div><div class="citation">IEEE Trans Vis Comput Graph. 2008 Nov-Dec; 14(6):1364-71.</div><p>[<a href="http://www.ncbi.nlm.nih.gov/pubmed/18988985/">PubMed</a>] [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R5">Ref list</a>]</p></div><div id="body-link-popper-R19" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">Learning to detect natural image boundaries using local brightness, color, and texture cues.<div class="authors">Martin DR, Fowlkes CC, Malik J</div><div class="citation">IEEE Trans Pattern Anal Mach Intell. 2004 May; 26(5):530-49.</div><p>[<a href="http://www.ncbi.nlm.nih.gov/pubmed/15460277/">PubMed</a>] [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R19">Ref list</a>]</p></div><div id="body-link-popper-R24" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">24. Reina G, Bidmon K, Enders F, Hastreiter P, Ertl T. GPU-Based Hyperstreamlines for Diffusion Tensor Imaging; Proceedings of EUROGRAPHICS - IEEE VGTC Symposium on Visualization 2006; 2006. pp. 35–42. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R24">Ref list</a>]</div><div id="body-link-popper-R23" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">Visualizing whole-brain DTI tractography with GPU-based Tuboids and LoD management.<div class="authors">Petrovic V, Fallon J, Kuester F</div><div class="citation">IEEE Trans Vis Comput Graph. 2007 Nov-Dec; 13(6):1488-95.</div><p>[<a href="http://www.ncbi.nlm.nih.gov/pubmed/17968101/">PubMed</a>] [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R23">Ref list</a>]</p></div><div id="body-link-popper-R27" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true"><span class="invert">Review</span> Circuit reconstruction tools today.<div class="authors">Smith SJ</div><div class="citation">Curr Opin Neurobiol. 2007 Oct; 17(5):601-8.</div><p>[<a href="http://www.ncbi.nlm.nih.gov/pubmed/18082394/">PubMed</a>] [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R27">Ref list</a>]</p></div><div id="body-link-popper-R9" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">Reconstruct: a free editor for serial section microscopy.<div class="authors">Fiala JC</div><div class="citation">J Microsc. 2005 Apr; 218(Pt 1):52-61.</div><p>[<a href="http://www.ncbi.nlm.nih.gov/pubmed/15817063/">PubMed</a>] [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R9">Ref list</a>]</p></div><div id="body-link-popper-R1" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">1. Anandan P. A computational framework and an algorithm for the measurement of visual motion. Journal on Computer Vision. 1989;2:283–310. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R1">Ref list</a>]</div><div id="body-link-popper-R6" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">6. Clarenz U, Droske M, Rumpf M. Inverse Problems, Image Analysis and Medical Imaging, AMS Special Session Interaction of Inverse Problems and Image Analysis. volume 313. AMS; 2002. Towards fast non–rigid registration; pp. 67–84. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R6">Ref list</a>]</div><div id="body-link-popper-R16" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">16. Lefohn A, Kniss J, Hansen C, Whitaker R. Interactive deformation and visualization of level set surfaces using graphics hardware. Proceedings of IEEE Visualization. 2003:75–82. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R16">Ref list</a>]</div><div id="body-link-popper-R26" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">26. Sethian J. Level set methods and fast marching methods. Cambridge University Press; 2002. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R26">Ref list</a>]</div><div id="body-link-popper-R12" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">12. Jeong W-K, Whitaker RT. A fast iterative method for Eikonal equations. SIAM Journal on Scientific Computing. 2008;30(5):2512–2534. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R12">Ref list</a>]</div><div id="body-link-popper-R31" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">31. Tomasi C, Manduchi R. Bilateral filtering for gray and color images. ICCV ’98. 1998:839–846. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R31">Ref list</a>]</div><div id="body-link-popper-R22" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">22. Perona P, Malik J. Scale space and edge detection using anisotropic diffusion. IEEE Trans. in Pattern Analysis and Machine Intelligence. 1990;volume 12:629–639. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R22">Ref list</a>]</div><div id="body-link-popper-R30" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">30. Tasdizen T, Whitaker R, Marc R, Jones B. Enhancement of cell boundaries in transmission micropscopy images; IEEE International Conf. on Image Processing (ICIP ’05); 2005. pp. 129–132. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R30">Ref list</a>]</div><div id="body-link-popper-R18" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">18. Maire M, Arbelaez P, Fowlkes C, Malik J. Using contours to detect and localize junctions in natural images; IEEE Conference on Computer Vision and Pattern Recognition (CVPR’08); 2008. pp. 1–8. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R18">Ref list</a>]</div><div id="body-link-popper-R10" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">10. Hadwiger M, Sigg C, Scharsach H, Bühler K, Gross M. Real-time ray-casting and advanced shading of discrete isosurfaces. Computer Graphics Forum (Proc. Eurographics 2005) 2005;24(3):303–312. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R10">Ref list</a>]</div><div id="body-link-popper-R8" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">8. Dice LR. Measures of the amount of ecologic association between species. Ecology. 1945;26:297–302. [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R8">Ref list</a>]</div><div id="body-link-popper-R29" class="body-link-popper ui-helper-reset ui-ncbipopper-wrapper ui-ncbipopper-basic" style="display: none; width: 30em; top: -100px; left: -100px;" aria-live="assertive" aria-hidden="true">MRI tissue classification with neighborhood statistics: a nonparametric, entropy-minimizing approach.<div class="authors">Tasdizen T, Awate SP, Whitaker RT, Foster NL</div><div class="citation">Med Image Comput Comput Assist Interv. 2005; 8(Pt 2):517-25.</div><p>[<a href="http://www.ncbi.nlm.nih.gov/pubmed/16685999/">PubMed</a>] [<a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#R29">Ref list</a>]</p></div></div>
308 <div id="NCBIFooter_dynamic">
309 <div class="breadcrumbs">You are here:
310 <span id="breadcrumb_text"><a href="http://www.ncbi.nlm.nih.gov/guide/">NCBI</a> > <a href="http://www.ncbi.nlm.nih.gov/guide/literature/">Literature</a> > <a href="http://www.ncbi.nlm.nih.gov/pmc/">PubMed Central (PMC)</a></span></div>
311 <a id="help-desk-link" class="help_desk jig-ncbihelpwindow" target="_blank" href="http://www.ncbi.nlm.nih.gov/sites/ehelp?&Ncbi_App=pmc&Db=pmc&Page=literature&Snapshot=/projects/PMC/PMCViewer@3.3&Time=2013-09-17T08:44:14-04:00&Host=ptpmc202">Write to the Help Desk</a>
315 <div class="footer" id="footer">
317 <div class="subfooter"> </div><script type="text/javascript" src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/preloaderWidget.js"> </script>
318 <div id="external-disclaimer" class="offscreen_noflow">
319 External link. Please review our <a href="http://www.nlm.nih.gov/privacy.html">privacy policy</a>.
321 <div id="ncbifooter" class="contact_info">
322 <div id="footer-contents-right">
323 <div id="nlm_thumb_logo">
324 <a href="http://www.nlm.nih.gov/" title="NLM">NLM</a>
326 <div id="nih_thumb_logo">
327 <a href="http://www.nih.gov/" title="NIH">NIH</a>
329 <div id="hhs_thumb_logo">
330 <a href="http://www.dhhs.gov/" title="DHHS">DHHS</a>
332 <div id="usagov_thumb_logo">
333 <a href="http://www.usa.gov/" title="USA.gov">USA.gov</a>
337 <div id="footer-contents-left">
339 <a href="http://www.ncbi.nlm.nih.gov/About/disclaimer.html">Copyright</a> | <a href="http://www.ncbi.nlm.nih.gov/About/disclaimer.html#disclaimer">Disclaimer</a> |
340 <a href="http://www.nlm.nih.gov/privacy.html" class="newdomain">Privacy</a> |
341 <a href="http://www.ncbi.nlm.nih.gov/guide/browsers/">Browsers</a> | <a href="http://www.nlm.nih.gov/accessibility.html">Accessibility</a> | <a href="http://www.ncbi.nlm.nih.gov/About/glance/contact_info.html">Contact</a>
342 <p class="address vcard">
344 <a class="fn url newdomain" href="http://www.ncbi.nlm.nih.gov/">National Center for
345 Biotechnology Information</a>,
346 </span> <span class="org url"><a href="http://www.nlm.nih.gov/">U.S. National Library of Medicine</a></span>
348 <span class="street-address">8600 Rockville Pike</span>, <span class="locality">Bethesda</span>
349 <span class="region">MD</span>, <span class="postal-code">20894</span>
350 <span class="country-name">USA</span>
355 <script type="text/javascript" src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/InstrumentPageStarterJS.js"> </script>
356 <script type="text/javascript" src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/hfjs.js"> </script>
363 </div><!-- /.twelve_col -->
364 <div class="ui-ncbiautocomplete-holder shadow" aria-live="assertive"><ul class="ui-ncbiautocomplete-options" role="listbox" aria-activedescendant=""></ul><div class="ui-ncbiautocomplete-actions shadow"><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" class="ui-ncbiautocomplete-link-pref">Preferences</a><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" class="ui-ncbiautocomplete-link-off">Turn off</a></div></div><div style="display: none;">External link. Please review our <a href="http://www.nlm.nih.gov/privacy.html">privacy policy</a>.</div></div>
367 <span class="PAFAppResources"></span>
369 <!-- BESelector tab -->
373 <noscript><img alt="statistics" src="/stat?jsdisabled=true&amp;ncbi_db=pmc&amp;ncbi_pdid=article&amp;ncbi_acc=&amp;ncbi_domain=nihpa&amp;ncbi_report=record&amp;ncbi_type=fulltext&amp;ncbi_objectid=&amp;ncbi_pcid=/articles/PMC3179915/&amp;ncbi_app=pmc" /></noscript>
376 <!-- usually for JS scripts at page bottom -->
377 <!--<component id="PageFixtures" label="styles"></component>-->
380 <!-- 8A1A29201DE7C981_0098SID /projects/PMC/PMCViewer@3.3 ptpmc202 v3.5.1.r392364: Mon, Mar 25 2013 15:07:09 -->
382 <script type="text/javascript" src="./Scalable and Interactive Segmentation and Visualization of Neural Processes in EM Datasets_files/3839767.js" snapshot="pmc"></script>
383 <div style="display: none; z-index: 1000; outline: 0px;" class="ui-dialog ui-widget ui-widget-content ui-corner-all" tabindex="-1" role="dialog" aria-labelledby="ui-dialog-title-epubDialog"><div class="ui-dialog-titlebar ui-widget-header ui-corner-all ui-helper-clearfix"><span class="ui-dialog-title" id="ui-dialog-title-epubDialog">Making articles easier to read in PMC</span><a href="http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179915/#" class="ui-dialog-titlebar-close ui-corner-all" role="button"><span class="ui-icon ui-icon-closethick">close</span></a></div><div id="epubDialog" style="" class="ui-dialog-content ui-widget-content"> <p>We are experimenting with display styles that make it easier to read articles in PMC. Our first effort uses eBook readers, which have several "ease of reading" features already built in.</p> <p>These PMC articles are best viewed in the <em>iBooks reader</em>. You may notice problems with the display of certain parts of an article in other eReaders.</p> <div class="ui-dialog-buttonpane ui-widget-content ui-helper-clearfix"><button id="cancelEpub" class="ui-state-default ui-corner-all">Cancel</button><button id="downloadEpub" style="float: left" class="ui-state-default ui-corner-all">Download article</button></div> </div></div><ul class="print-log"><li></li></ul><div id="MathJax_Font_Test" style="position: absolute; visibility: hidden; top: 0px; left: 0px; width: auto; padding: 0px; border: 0px; margin: 0px; white-space: nowrap; text-align: left; text-indent: 0px; text-transform: none; line-height: normal; letter-spacing: normal; word-spacing: normal; font-size: 40px; font-weight: normal; font-style: normal; font-family: MathJax_Size2, sans-serif;"></div></body></html>