Exploiting Deep Learning Techniques for Colon Polyp Segmentation

• DOI: 10.32604/cmc.2021.013618
• Publication Date: 2021-02-23T02:30:14Z
• Authors: Daniel Sierra-Sosa; Sebastian Patino-...; Begonya Garcia-Za...; Cristian Castillo...; Adel Elmaghraby
• Keywords: 03 medical and health sciences; 0302 clinical medicine

As colon cancer is among the top causes of death, there a growing interest in developing improved techniques for early detection polyps. Given close relation between polyps and cancer, their helps avoid cases. The increment availability colorectal screening tests number colonoscopies have increased burden on medical personnel. In this article, application deep learning segmentation presented. Four were implemented evaluated: Mask-RCNN, PANet, Cascade R-CNN Hybrid Task (HTC). These trained tested using CVC-Colon database, ETIS-LARIB Polyp, proprietary dataset. Three experiments conducted to assess performance: 1) Training testing each database independently, 2) Mergingd databases independently merged test set, 3) dataset set. our experiments, PANet architecture has best performance Polyp detection, HTC was most accurate segment them. This approach allows us employ Deep Learning assist healthcare professionals diagnosis cancer. It anticipated that can be part framework semi-automated polyp colonoscopies.