Polyp segmentation and identification from endoscopic images by implementing deep learning models

Abstract

Colorectal cancer (CRC) is one of the leading causes of cancer-related mortality worldwide, with early detection of colorectal polyps playing a vital role in reducing its impact. Manual identification of polyps during colonoscopy is often limited by human error, variability in polyp appearance, and real-time diagnostic constraints. This work explores a deep learning-based approach to automatic polyp segmentation in endoscopic images, aiming to enhance detection accuracy and reduce diagnostic delays. Multiple convolutional and transformer-based segmentation models are implemented to learn pixel-level features associated with polyp structures. The publicly available Kvasir-SEG dataset is utilized and further augmented to improve training diversity. A specialized boundary-aware loss function is introduced to address the challenge of ambiguous polyp borders. The proposed method emphasizes the importance of robust preprocessing, architectural design, and edge-focused optimization to advance the reliability of computer-aided diagnosis systems in clinical gastroenterology.