Deep Learning-Based Comparative Study to Detect Polyp Removal in Endoscopic Images

Polyps are one of the most common gastrointestinal diseases. It has the potential to cause fatal colon and rectal cancers. As a result, it must be removed during the primitive stage. In this paper, we developed an algorithm that uses endoscopy images to detect polyp removal status. We investigated convolutional neural networks such as DenseNet, ResNet, VGG, MobileNet, and others to extract features from images and then use those features to classify whether a polyp is completely removed or not. 1000 dyed resection margins and 1000 dyed and lifted polyps' images from a publicly available dataset were used to test and train the proposed models. On the testing dataset, we obtained 85% sensitivity, 88% precision, and 85% fl-scores by using MobileNet architecture. This computer-aided polyp removal method assists physicians in diagnosing polyp status in a reliable, quick, and cost-effective manner.