-
Notifications
You must be signed in to change notification settings - Fork 0
/
index.html
executable file
·189 lines (164 loc) · 9.49 KB
/
index.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
<!DOCTYPE html>
<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<meta name="viewport" content="width=device-width,initial-scale=1">
<title>Self-similarity Student for Partial Label Histopathology Image Segmentation</title>
<link href="css/bootstrap.min.css" rel="stylesheet">
<link rel="stylesheet" href="css/main.css">
<script src='https://code.jquery.com/jquery-2.2.4.min.js'></script>
<meta name="description" content="Project Website">
</head>
<body class="nd-docs">
<div class="nd-pageheader">
<div class="container">
<h1>Self-similarity Student for Partial Label <br> Histopathology Image Segmentation</h1>
<address>
<a href="">Hsien-Tzu Cheng<sup>*,1</sup></a>,
<a href="">Chun-Fu Yeh<sup>*,1</sup></a>,
<a href="">Po-Chen Kuo<sup>1,3</sup></a>,
<a href="">Andy Wei<sup>1</sup></a>,
<a href="">Keng-Chi Liu<sup>1</sup></a>,
<a href="">Mong-Chi Ko<sup>1</sup></a>, <br>
<a href="">Kuan-Hua Chao<sup>1</sup></a>,
<a href="">Yu-Ching Peng<sup>2</sup></a>,
<a href="">Tyng-Luh Liu<sup>1,4</sup></a>
<br><br>
<nobr>Taiwan AI Labs<sup>1</sup></nobr>,
<nobr>Taipei Veterans General Hospital<sup>2</sup></nobr>, <br>
<nobr>National Taiwan University College of Medicine<sup>3</sup></nobr>, <br>
<nobr>Institute of Information Science, Academia Sinica, Taiwan<sup>4</sup></nobr> <br>
<nobr>(* Both authors contributed equally to this work.)</nobr>
</address>
</div>
<a href="https://arxiv.org/abs/2007.09610" target="_blank"><img src="img/pdf.png" style="width:40px;height:40px;"> [Paper (arXiv)] </a>
<a href="https://www.youtube.com/embed/cyqpIton2Oo" target="_blank"><img src="img/youtube.png" style="width:40px;height:40px;"> [Video] </a>
<p></p>
</div><!-- end nd-pageheader -->
<div class="container">
<div class="row">
<div class="col">
<figure>
<img src="img/fig_overview_v15-min.png" style="width:640px;">
</figure>
<h2>Abstract</h2>
<p>Delineation of cancerous regions in gigapixel whole slide images (WSIs) is a crucial diagnostic procedure in digital pathology. This process is time-consuming because of the large search space in the gigapixel WSIs, causing chances of omission and misinterpretation at indistinct tumor lesions. To tackle this, the development of an automated cancerous region segmentation method is imperative. We frame this issue as a modeling problem with partial label WSIs, where some cancerous regions may be misclassified as benign and vice versa, producing patches with noisy labels. To learn from these patches, we propose Self-similarity Student, combining teacher-student model paradigm with similarity learning. Specifically, for each patch, we first sample its similar and dissimilar patches according to spatial distance. A teacher-student model is then introduced, featuring the exponential moving average on both student model weights and teacher predictions ensemble. While our student model takes patches, teacher model takes all their corresponding similar and dissimilar patches for learning robust representation against noisy label patches. Following this similarity learning, our similarity ensemble merges similar patches’ ensembled predictions as the pseudo-label of a given patch to counteract its noisy label. On the CAMELYON16 dataset, our method substantially outperforms state-of-the-art noise-aware learning methods by 5% and the supervised-trained baseline by 10% in various degrees of noise. Moreover, our method is superior to the baseline on our TVGH TURP dataset with 2% improvement, demonstrating the generalizability to more clinical histopathology segmentation tasks.</p>
<hr>
<!--figure>
<img src="img/preview.jpg">
</figure-->
<p></p>
<!--a href="" target="_blank"><img src="img/github_icon.png"> [Source Code] </a><p></p-->
<!--hr-->
</div>
</div> <!-- row -->
<div class="row">
<div class="col">
<h2>Method Overview</h2>
</div>
</div> <!-- row -->
<div class="row">
<iframe class='center' width="640" height="360" src="https://www.youtube.com/embed/cyqpIton2Oo" frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>
</div> <!-- row -->
<!--div class="row">
<div class="col">
<h2>Our Method</h2>
<figure>
<img src="img/fig_model_overview_v12-min.png" style="width:640px;">
<figcaption>
</figcaption>
</figure>
</div>
</div--> <!-- row -->
<br>
<br>
<div class="row">
<div class="col">
<h2>Result</h2>
</div>
</div>
<div class="row">
<div class="col">
<p style="text-align:center;">CAMELYON16 Result (trained from Top-1 setting with 42.4% wrong label)</p>
<div class='center' style="position:relative;max-width:720px;min-width:60px;height:480px;">
<iframe style="top: 0; left: 0; width: 100%; height: 100%;" src="https://flickrembed.com/cms_embed.php?source=flickr&layout=responsive&input=https%3A%2F%2Fwww.flickr.com%2Fphotos%2F189589104%40N02%2Falbums%2F72157715372938547&sort=0&by=album&theme=default_notextpanel&scale=fill&speed=3000&limit=10&skin=default&autoplay=false" scrolling="no" frameborder="0" allowFullScreen="true" webkitallowfullscreen="true" mozallowfullscreen="true"></iframe>
<script type="text/javascript">function showpics(){var a=$("#box").val();$.getJSON("http://api.flickr.com/services/feeds/photos_public.gne?tags="+a+"&tagmode=any&format=json&jsoncallback=?",function(a){$("#images").hide().html(a).fadeIn("fast"),$.each(a.items,function(a,e){$("<img/>").attr("src",e.media.m).appendTo("#images")})})}</script>
<div style="position: absolute;width: 80%;bottom:17px;left: 0;right: 0;margin-left: auto;margin-right: auto;color: #000;text-align: center;">
<!--small style="line-height:1.8;font-size:2px;background:transparent;">Powered by
<a href="https://flickrembedslideshow.com/fr/">Flickrembedslideshow.com/fr/</a> & <a href=""></a>
</small-->
</div>
</div>
</div>
</div>
<div class="row">
<div class="col">
<p style="text-align:center;">TVGH TURP Result (trained from all label)</p>
<div class='center' style="position:relative;max-width:720px;min-width:100px;height:480px;">
<iframe style="top: 0; left: 0; width: 100%; height: 100%;" src="https://flickrembed.com/cms_embed.php?source=flickr&layout=responsive&input=https%3A%2F%2Fwww.flickr.com%2Fphotos%2F189589104%40N02%2Falbums%2F72157715373392812&sort=0&by=album&theme=default_notextpanel&scale=fill&speed=3000&limit=10&skin=default&autoplay=false" scrolling="no" frameborder="0" allowFullScreen="true" webkitallowfullscreen="true" mozallowfullscreen="true"></iframe>
<script type="text/javascript">function showpics(){var a=$("#box").val();$.getJSON("http://api.flickr.com/services/feeds/photos_public.gne?tags="+a+"&tagmode=any&format=json&jsoncallback=?",function(a){$("#images").hide().html(a).fadeIn("fast"),$.each(a.items,function(a,e){$("<img/>").attr("src",e.media.m).appendTo("#images")})})}</script>
<div style="position: absolute;width: 80%;bottom:17px;left: 0;right: 0;margin-left: auto;margin-right: auto;color: #000;text-align: center;">
<!--small style="line-height:1.8;font-size:2px;background:transparent;">Powered by
<a href="https://flickrembedslideshow.com/de/">Flickrembedslideshow.com/de/</a> & <a href=""></a>
</small-->
</div>
</div>
</div>
</div>
<!--div class="row">
<div class="col">
<div class="img-zoom-container">
<img id="myimage" src="img/S108-76745J-prob.jpg" width="500" height="300" alt="WSI">
</div>
</div>
<div class="col">
<div id="myresult" class="img-zoom-result"></div>
</div>
</div--> <!-- row -->
</div>
<div class="container" id="paper">
<div class="row">
<div class="col">
<h2>Citation</h2>
<pre>@InProceedings{Cheng_2020_ECCV,
author = {Cheng, Hsien-Tzu and Yeh, Chun-Fu and Kuo, Po-Chen and Wei, Andy and Liu, Keng-Chi and Ko, Mong-Chi and Chao, Kuan-Hua and Peng, Yu-Ching and Liu, Tyng-Luh},
title = {Self-similarity Student for Partial Label Histopathology Image Segmentation},
booktitle = {Proceedings of the European Conference on Computer Vision (ECCV)},
year = {2020}
}</pre>
</div>
</div>
</div>
<div class="container">
<div class="row">
<div class="col">
<h2>Acknowledgement</h2>
We appreciate Yi-Chin Tu, the chairman of Taiwan AI Labs, for the generous support of this project. <br>
We appreciate the intensive assistance made by the Department of Pathology and Laboratory Medicine, Taipei Veterans General Hospital. <br>
We appreciate Tsun-Hsiao Wang at National Yang-Ming University for his contribution on cancer delineation in WSIs of TVGH TURP dataset.
</div>
</div> <!-- row -->
</div> <!-- container -->
<br>
<div class="container">
<div class="row">
<div class="col">
<p style="text-align:center;">
<a href="" target="_blank"><img src="img/ailabs_logo.png" height="60"></a>
 
<a href="" target="_blank"><img src="img/tvgh_logo_big.png" height="90"></a>
 
</p>
</div>
</div>
</div>
</div>
<div class="footer">
<p>© 2020 Copyright Taiwan AI Labs. All Rights Reserved. </p>
</div>
<!--script src='js/main.js'></script-->
<!--script>
imageZoom("myimage", "myresult");
</script-->
</body>
</html>