Attention Guided U-Net with Grad CAM for Explainable Polyp Segmentation in Colonoscopy Images

Colorectal cancer is one of the leading causes of cancer-related deaths worldwide, and early detection through colonoscopy is essential for effective treatment. This project presents an innovative approach for polyp segmentation in colonoscopy images using deep learning techniques. We leverage attention-guided U-Net models, including Swin-CNN, ViT-U-Net, and U-Net++, to enhance the accuracy and efficiency of polyp detection. In addition, Grad-CAM (Gradient-weighted Class Activation Mapping) is integrated into the framework to provide explainable AI, allowing clinicians to interpret the model's decision-making process. The entire system is deployed as a web application built with Flask, HTML, and CSS, where users can log in, register, and upload colonoscopy images for segmentation prediction. The model outputs the predicted mask, while Grad-CAM visualizations highlight the regions of the image that contribute most to the model’s decision. This user-friendly interface aims to assist healthcare professionals in diagnosing and monitoring colorectal conditions more effectively, ensuring a higher level of trust and transparency in AI-powered medical tools.

Keywords:

Polyp Segmentation, Colonoscopy, Deep Learning, Attention-Guided U-Net, Swin-CNN, ViT-U-Net, U-Net++, Grad-CAM, Explainable AI, Flask Web Application.