{"title":"A Practical Polyp Detecting Model in Colonoscopy Video by Post-processing","authors":"Zhipeng Zhang, Ling Ma, Y-D. Chana, Li Xiao, Qing He, Conghui Ma, Huiqin Jiang","doi":"10.1109/CISP-BMEI53629.2021.9624346","DOIUrl":null,"url":null,"abstract":"Polyp recognition in colonoscopy video is crucial for early colorectal cancer detection and treatment. However, polyps are very similar to other intestinal tissues. Intestinal peristalsis and debris shelter lead to great changes in polyp morphology. Lens motion may lead to blurred images. Traditional target detection methods can not meet the needs of the complex environment. In previous work, we propose top likelihood loss and similarity loss to solve the false positive problem. However, when detecting video polyps, there are great problems in the previous work due to the more complex video environment. In this work, we develop the new video detection mode based on our previous work. We add a new post-processing method in the prediction part of YOLOv4 model, which uses the front and rear neighborhood frames to judge the accuracy of current frame detection, combine the single frame detection results and spatiotemporal information, and then make the final decision. We test our method on two datasets, One is the private dataset we collect, and the other is the public dataset. Compared with the baseline, our method has a great improvement in accuracy. Ours method is superior to the most advanced target detection methods that can meet real-time constraints.","PeriodicalId":131256,"journal":{"name":"2021 14th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)","volume":"7 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2021-10-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2021 14th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/CISP-BMEI53629.2021.9624346","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
Polyp recognition in colonoscopy video is crucial for early colorectal cancer detection and treatment. However, polyps are very similar to other intestinal tissues. Intestinal peristalsis and debris shelter lead to great changes in polyp morphology. Lens motion may lead to blurred images. Traditional target detection methods can not meet the needs of the complex environment. In previous work, we propose top likelihood loss and similarity loss to solve the false positive problem. However, when detecting video polyps, there are great problems in the previous work due to the more complex video environment. In this work, we develop the new video detection mode based on our previous work. We add a new post-processing method in the prediction part of YOLOv4 model, which uses the front and rear neighborhood frames to judge the accuracy of current frame detection, combine the single frame detection results and spatiotemporal information, and then make the final decision. We test our method on two datasets, One is the private dataset we collect, and the other is the public dataset. Compared with the baseline, our method has a great improvement in accuracy. Ours method is superior to the most advanced target detection methods that can meet real-time constraints.