Gastric Polyp Detection Using Deep Convolutional Neural Network

Abstract

Certain types of gastric polyps may increase the risk of stomach cancer. With recent progress in computer vision due to deep learning, it is possible to reduce the gastric polyp miss rate, leading to a better and more accurate endoscopy. Automating the process of gastric polyp detection is a complex task as polyps differ in terms of size, shape and texture. Yolov3 is a fast and accurate object detection algorithm. In this paper, yolov3 is used for polyp detection. The results show that yolov3 used for gastric polyp detection can achieve a mean average precision (mAP) of 0.91. Also, yolov3-tiny, a smaller and faster version achieves a mean average precision (mAP) of 0.82 and gives more than 100 fps on Pascal Titan X GPU. This can help endoscopic physicians increase productivity.