Unet and attention network-based deep learning technique for polyp segmentation

Abstract

Colon cancer is one of the leading causes of cancer-related deaths worldwide, and early detection of precancerous polyps is crucial for effective treatment. Polyp segmentation has accomplished massive triumph over the years in supervised learning. However, obtaining many labeled datasets is commonly challenging in the medical domain. To solve this problem, we developed a fully automated pixel-wise polyp segmentation model using the U-Net. The U-Net architecture is trained on Kvasir-SEG and CVC-ClinicDB, two open-access datasets of gastrointestinal polyp images and corresponding segmentation masks, manually annotated and verified by an experienced gastroenterologist. The network is designed to predict a pixel-wise segmentation mask of the polyp region in the input image. We demonstrate that the U-Net model achieves high accuracy in polyp segmentation. This paper also developed an extension of the U-Net architecture for polyp segmentation that incorporates an attention mechanism. During the segmentation process, the attention mechanism is used to selectively highlight relevant regions of the colonoscopy image, such as the polyp region. We demonstrate that the Attention U-Net model achieves higher accuracy in polyp segmentation than the original U-Net. The attention mechanism also provides insights into the importance of different regions of the colonoscopy image for accurate polyp segmentation. We believe this approach can improve colon cancer diagnosis and treatment accuracy and efficiency.