{"id":3407,"date":"2023-08-31T14:53:45","date_gmt":"2023-08-31T19:53:45","guid":{"rendered":"https:\/\/my.vanderbilt.edu\/masi\/?p=3407"},"modified":"2023-08-31T14:53:45","modified_gmt":"2023-08-31T19:53:45","slug":"semantic-aware-contrastive-learning-for-multi-object-medical-image-segmentation","status":"publish","type":"post","link":"https:\/\/my.vanderbilt.edu\/masi\/2023\/08\/semantic-aware-contrastive-learning-for-multi-object-medical-image-segmentation\/","title":{"rendered":"Semantic-Aware Contrastive Learning for Multi-object Medical Image Segmentation"},"content":{"rendered":"<p>Ho Hin Lee, Yucheng Tang, Qi Yang, Xin Yu, Leon Y. Cai, Lucas W. Remedios, Shunxing Bao, Bennett A. Landman, Yuankai Huo<\/p>\n<p>Paper:\u00a0<a href=\"https:\/\/ieeexplore.ieee.org\/document\/10149329\">https:\/\/ieeexplore.ieee.org\/document\/10149329<\/a><\/p>\n<p>Code:\u00a0<a href=\"https:\/\/github.com\/MASILab\/DCC_CL\">https:\/\/github.com\/MASILab\/DCC_CL<\/a><\/p>\n<h2>Abstract<\/h2>\n<p>Medical image segmentation, or computing voxel-wise semantic masks, is a fundamental yet challenging task in medical imaging domain. To increase the ability of encoder-decoder neural networks to perform this task across large clinical cohorts, contrastive learning provides an opportunity to stabilize model initialization and enhances downstream tasks performance without ground-truth voxel-wise labels. However, multiple target objects with different semantic meanings and contrast level may exist in a single image, which poses a problem for adapting traditional contrastive learning methods from prevalent \u201cimage-level classification\u201d to \u201cpixel-level segmentation\u201d. In this paper, we propose a simple semantic-aware contrastive learning approach leveraging attention masks and image-wise labels to advance multi-object semantic segmentation. Briefly, we embed different semantic objects to different clusters rather than the traditional image-level embeddings. We evaluate our proposed method on a multi-organ medical image segmentation task with both in-house data and MICCAI Challenge 2015 BTCV datasets. Compared with current state-of-the-art training strategies, our proposed pipeline yields a substantial improvement of 5.53% and 6.09% on Dice score for both medical image segmentation cohorts respectively (p-value &lt; 0.01). The performance of the proposed method is further assessed on external medical image cohort via MICCAI Challenge FLARE 2021 dataset, and achieves a substantial improvement from Dice 0.922 to 0.933 (p-value &lt; 0.01). The code is available at: https:\/\/github.com\/MASILab\/DCC_CL<\/p>\n<p><a href=\"https:\/\/cdn.vanderbilt.edu\/vu-my\/wp-content\/uploads\/sites\/2304\/2023\/08\/31145121\/Screenshot-2023-08-31-at-2.50.56-PM.png\"><img loading=\"lazy\" decoding=\"async\" src=\"https:\/\/my.vanderbilt.edu\/masi\/wp-content\/uploads\/sites\/2304\n2661\/2023\/08\/Screenshot-2023-08-31-at-2.50.56-PM-650x371.png\" alt=\"Screenshot 2023-08-31 at 2.50.56 PM\" width=\"650\" height=\"371\" class=\"alignnone size-large wp-image-3408\" srcset=\"https:\/\/cdn.vanderbilt.edu\/t2-my\/my-prd\/wp-content\/uploads\/sites\/2304\/2023\/08\/Screenshot-2023-08-31-at-2.50.56-PM-650x371.png 650w, https:\/\/cdn.vanderbilt.edu\/t2-my\/my-prd\/wp-content\/uploads\/sites\/2304\/2023\/08\/Screenshot-2023-08-31-at-2.50.56-PM-300x171.png 300w, https:\/\/cdn.vanderbilt.edu\/t2-my\/my-prd\/wp-content\/uploads\/sites\/2304\/2023\/08\/Screenshot-2023-08-31-at-2.50.56-PM-768x439.png 768w\" sizes=\"auto, (max-width: 650px) 100vw, 650px\" \/><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Ho Hin Lee, Yucheng Tang, Qi Yang, Xin Yu, Leon Y. Cai, Lucas W. Remedios, Shunxing Bao, Bennett A. Landman, Yuankai Huo Paper:\u00a0https:\/\/ieeexplore.ieee.org\/document\/10149329 Code:\u00a0https:\/\/github.com\/MASILab\/DCC_CL Abstract Medical image segmentation, or computing voxel-wise semantic masks, is a fundamental yet challenging task in medical imaging domain. To increase the ability of encoder-decoder neural networks to perform this task&#8230;<\/p>\n","protected":false},"author":1920,"featured_media":3408,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[5,64,60,130,116,3,114,23],"tags":[],"class_list":["post-3407","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-abdomen-imaging","category-body-wise","category-computed-tomography","category-deep-learning","category-image-processing","category-image-segmentation","category-labeling","category-machine-learning"],"_links":{"self":[{"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/posts\/3407","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/users\/1920"}],"replies":[{"embeddable":true,"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/comments?post=3407"}],"version-history":[{"count":1,"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/posts\/3407\/revisions"}],"predecessor-version":[{"id":3409,"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/posts\/3407\/revisions\/3409"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/media\/3408"}],"wp:attachment":[{"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/media?parent=3407"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/categories?post=3407"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/my.vanderbilt.edu\/masi\/wp-json\/wp\/v2\/tags?post=3407"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}