Classification of Polyps and Adenomas Using Deep Learning Model in Screening Colonoscopy

Abstract

Colorectal cancer (CRC) is the third leading cause of cancer-related death in China. It usually originates from the non-cancerous neoplasm polyps of the colon or rectal epithelium. Some polyps will evolve into precancerous lesions and eventually turn into colorectal cancer, Early screening and removal of adenomas can reduce the risk of colorectal cancer if screened. Unfortunately, more than 60% of colorectal cancer cases are attributed to missed polyps. Therefore, a deep learning network referred to as the faster_rcnn_inception_ resnet_v2 model was introduced for the localization and classification of precancerous lesions. It enables high-precision classification of polyps and adenomas under white light endoscopic images. The Mean Average Precision reached 90.645% when the Intersection over Union is set to 0.5. As an aid to clinicians, the model can improve the detection rate of adenomas and the diagnostic accuracy of early CRC.