{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/recurrent-saliency-transformation-network","title":"Recurrent Saliency Transformation Network: Incorporating Multi-Stage Visual Cues for Small Organ Segmentation","arxiv_id":"1709.04518","date":"2017-09-13","proceeding":"CVPR 2018 6","authors":["Qihang Yu","Lingxi Xie","Yan Wang","Yuyin Zhou","Elliot K. Fishman","Alan L. Yuille"],"abstract":"We aim at segmenting small organs (e.g., the pancreas) from abdominal CT\nscans. As the target often occupies a relatively small region in the input\nimage, deep neural networks can be easily confused by the complex and variable\nbackground. To alleviate this, researchers proposed a coarse-to-fine approach,\nwhich used prediction from the first (coarse) stage to indicate a smaller input\nregion for the second (fine) stage. Despite its effectiveness, this algorithm\ndealt with two stages individually, which lacked optimizing a global energy\nfunction, and limited its ability to incorporate multi-stage visual cues.\nMissing contextual information led to unsatisfying convergence in iterations,\nand that the fine stage sometimes produced even lower segmentation accuracy\nthan the coarse stage.\n  This paper presents a Recurrent Saliency Transformation Network. The key\ninnovation is a saliency transformation module, which repeatedly converts the\nsegmentation probability map from the previous iteration as spatial weights and\napplies these weights to the current iteration. This brings us two-fold\nbenefits. In training, it allows joint optimization over the deep networks\ndealing with different input scales. In testing, it propagates multi-stage\nvisual information throughout iterations to improve segmentation accuracy.\nExperiments in the NIH pancreas segmentation dataset demonstrate the\nstate-of-the-art accuracy, which outperforms the previous best by an average of\nover 2%. Much higher accuracies are also reported on several small organs in a\nlarger dataset collected by ourselves. In addition, our approach enjoys better\nconvergence properties, making it more efficient and reliable in practice.","url_abs":"http://arxiv.org/abs/1709.04518v4","url_pdf":"http://arxiv.org/pdf/1709.04518v4.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"recurrent-saliency-transformation-network","repo_url":"https://github.com/198808xc/OrganSegRSTN","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"caffe2","reach":null},{"paper_slug":"recurrent-saliency-transformation-network","repo_url":"https://github.com/twni2016/OrganSegRSTN_PyTorch","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"organ-segmentation","task_name":"Organ Segmentation"},{"task_slug":"pancreas-segmentation","task_name":"Pancreas Segmentation"},{"task_slug":"segmentation","task_name":"Segmentation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/pancreas-segmentation-on-tcia-pancreas-ct","task":"Pancreas Segmentation","dataset":"TCIA Pancreas-CT Dataset","model":"Recurrent Saliency Transformation Network","rank_in_archive_order":1,"of":3,"metrics":{"Dice Score":"0.845"},"uses_additional_data":false}],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}