Advancing Colorectal Polyp Detection and Segmentation Through YOLOv11 Architecture

Timely and precise detection of colorectal polyps plays a vital role in the early diagnosis and prevention of colorectal cancer, one of the leading causes of cancer-related mortality worldwide. This study introduces a robust deep learning framework based on the state-of-the-art YOLOv11 architecture for real-time polyp detection and segmentation in colonoscopy images. Leveraging the Kvasir-SEG dataset for both training and evaluation, the proposed model incorporates key architectural enhancements, including Cross-Stage Partial Spatial Attention (C2PSA), resolution-preserving pathways, and dynamic multitask loss balancing to improve object localization and boundary precision. Experiments were conducted across multiple variants of YOLOv11-nano, small, medium, and large-and performance was assessed using standard evaluation metrics such as precision, recall, mAP, IoU, Dice and F1-score. The results demonstrate that YOLOv11 consistently outperforms prior YOLO versions and other leading detection and segmentation models, particularly in identifying small and morphologically complex polyps. Visual assessments further validate the model’s capability to produce high-confidence, pixel-accurate predictions. Overall, the YOLOv11-based system offers an effective, scalable, and clinically viable solution for automated colorectal polyp analysis in computer-aided endoscopic diagnosis.