2021/11/08 by Jiacheng Wang, Wang, Jiacheng, Yueming Jin +11
Medicine · #Colorectal Cancer Screening and Detection #Computer Vision and Pattern Recognition (cs.CV) #Esophageal Cancer Research and Treatment #FOS: Computer and information sciences #FOS: Electrical engineering #Gastric Cancer Management and Outcomes #Image and Video Processing (eess.IV) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2111.04733
openalex publication_date 2021/11/08 · openalex created_date 2022/05/05 · openalex updated_date 2026/07/28
We propose a novel shape-aware relation network for accurate and real-time\nlandmark detection in endoscopic submucosal dissection (ESD) surgery. This task\nis of great clinical significance but extremely challenging due to bleeding,\nlighting reflection, and motion blur in the complicated surgical environment.\nCompared with existing solutions, which either neglect geometric relationships\namong targeting objects or capture the relationships by using complicated\naggregation schemes, the proposed network is capable of achieving satisfactory\naccuracy while maintaining real-time performance by taking full advantage of\nthe spatial relations among landmarks. We first devise an algorithm to\nautomatically generate relation keypoint heatmaps, which are able to\nintuitively represent the prior knowledge of spatial relations among landmarks\nwithout using any extra manual annotation efforts. We then develop two\ncomplementary regularization schemes to progressively incorporate the prior\nknowledge into the training process. While one scheme introduces pixel-level\nregularization by multi-task learning, the other integrates global-level\nregularization by harnessing a newly designed grouped consistency evaluator,\nwhich adds relation constraints to the proposed network in an adversarial\nmanner. Both schemes are beneficial to the model in training, and can be\nreadily unloaded in inference to achieve real-time detection. We establish a\nlarge in-house dataset of ESD surgery for esophageal cancer to validate the\neffectiveness of our proposed method. Extensive experimental results\ndemonstrate that our approach outperforms state-of-the-art methods in terms of\naccuracy and efficiency, achieving better detection results faster. Promising\nresults on two downstream applications further corroborate the great potential\nof our method in ESD clinical practice.\n