Real-Time Detection of Colorectal Adenomas Based on an Enhanced YOLOv5 Network

Colorectal cancer remains one of the leading causes of morbidity and mortality worldwide, posing a serious threat to human health.Colonoscopy is considered the gold standard for colorectal cancer diagnosis, and adenoma detection rate (ADR) is a critical quality indicator of colonoscopic procedures.However, accurate detection is challenging due to the complex intestinal structure under colonoscopy, the presence of interfering elements such as foam, residue, and feces, and the diverse sizes and subtle features of adenomatous polyps.Moreover, most existing AI-based detection models are computationally intensive and too slow for real-time application.To address these challenges, this study proposes a real-time colorectal adenoma detection method based on an improved single-stage object detection network, YOLOv5.To mitigate issues like false negatives and false positives in intermediate video frames caused by motion blur or camera defocus, the original algorithm is enhanced by integrating motion information and a sequential bounding-box matching post-processing module.This improves recall in video-based detection by correcting missed detections in intermediate frames.Performance validation on the public ImageNet VID dataset shows a 5.93% increase in mAP, while experiments on colonoscopy video datasets demonstrate a 20.47% improvement in recall rate.