Colonoscopy Polyp Segmentation using Deep Residual U-Net with Bottleneck Attention Module

The colonoscopy is the most reliable method for monitoring the digestive tract. Colonography can detect a variety of conditions, including polyps in the colon. Despite advancements in technology, many colorectal polyps still go undetected in the early stages. When polyps are detected at an early stage, the severity of the disease can be mitigated with the use of polyp segmentation. Coherence transfer and contrast-limited adaptive histogram equalization were two of the image pre-processing approaches used by the researchers in this work to address these issues. Following this, a U-Net based deep learning segmentation model was utilized to isolate the polyp in the image. Using a bottleneck attention module and a residual network, the BAMRes encoder-decoder component of the Unet framework's architecture is combined with feature concatenation on the same layer. With the publicly accessible Kvasir-SEG dataset, we were able to empirically validate the model, which yielded a dice coefficient of 92.27%.