Massive Colonoscopy Images Oriented Polyp Detection

Abstract

Since more than 90% of colorectal cancers are converted from colorectal polyps, colonoscopy is the most effective method for early detection of colorectal polyps. However, artificial polyp judgement leads to a high missed diagnosis rate during colonoscopy inspection. To reduce the missed diagnosis rate, we propose an end-to-end deep learning based polyp detection method combining a series of pretreatment methods with a multiple classification based detection network. We have compared our method with several currently popular object detection methods. Experiment results show that our method has effective improvements on detection precision and performance.