An approach of polyp segmentation from colonoscopy images using Dilated-U-Net-Seg – A deep learning network

Abstract

Throughout the entire world, colorectal cancer (CRC) is the second most common cause of death. The most important element in preventing cancer is early detection. It increases the likelihood that those in need will survive. There are several approaches for diagnosing CRC, but colonoscopy is still the go-to procedure for both screening and diagnosis. It finds polyps and other abnormal CRC indications in the colon and rectum. Although several technical developments have enhanced the early identification of polyps, there is still a considerable likelihood that polyps may be overlooked due to several variables. Techniques based on artificial intelligence are crucial for finding such polyps. This study develops the Dilated-U-Net-Seg Deep Learning-based technique for effectively segmenting the polyp area. The suggested framework combines dilated convolutions and feature concatenation in the same layer with an encoder-decoder structure. Five datasets have been divided into separate sets for training and testing is used to empirically test this framework. This yields 97.11 % of pixel accuracy, 97.90 % of dice score, 88 % of the Jaccard Index, 95.89 % of recall, and 92.40 % of precision. In comparison to the current U-Net architecture, this novel framework for polyp segmentation that implements Dilated-U-Net-Seg demonstrates a 3.05 % increase in Pixel accuracy.