Using YOLOv3 to detect colorectal polyps in video

Abstract

A model based on the You Only Look Once (YOLO) algorithm was trained to detect polyps on a dataset of 503 polyp-positive images taken from colonoscopy video frames. We achieve precision on test sets that is comparable to the performance of a previously described model trained on a much larger set of 5,545 images. However, our model has a lower recall and missed some polyps in the test set.