Analysis of Polyp Segmentation in Colonoscopy Imaging: A Comparative Study with U2Net

There is a growing interest in diagnostic processes using machine vision and deep learning techniques in medical imaging. This study contributes to this field by systematically evaluating various neural network architectures on four distinct datasets: ETIS-Larib, CVC Clinic DB, CVC Colon DB, and Kvasir-Seg. We compare the performance of models like U-Net, ResUNet, ResUNet++, RAUNet, DenseUNet, and U2-Net. Additionally, we provide a detailed analysis of the U2- Net architecture, highlighting its features and strong performance in polyp segmentation. Our results underscore the superiority of the U2-Net model, which consistently outperforms its counterparts across all datasets.