Publication

Abstract : Colonoscopy plays a pivotal role in detecting and diagnosing colorectal diseases, with polyp segmentation being a critical step for accurate diagnosis. In this study, we propose a novel approach for polyp segmentation in colonoscopy images, leveraging the Shuffle attention mechanism within the proposed architecture. Our method is rigorously evaluated across three diverse colonoscopy datasets, demonstrating promising results with an mean dice score of 0.93. Furthermore, to enhance segmentation accuracy, we employ a Conditional Random Field (CRF) post-processing method to refine the segmentation results. Through extensive experimentation and analysis, we showcase the effectiveness of our approach in achieving highly accurate polyp segmentation, thereby contributing to improved diagnostic outcomes in colorectal healthcare. Our method holds significant potential for enhancing computer-aided detection systems in clinical practice, facilitating early detection and treatment of colorectal abnormalities.