Precision diagnosis by integration of transfer learning for colorectal cancer polyp detection and classification

Abstract

Colorectal cancer (CRC) ranks among the most common and deadly cancers worldwide, posing a serious public health challenge. Fortunately, early detection significantly boosts the chances of successful treatment. Most CRC cases originate from adenomatous polyps—non-cancerous growths in the colon or rectum that can gradually transform into malignant tumours over time. Identifying  these polyps during colonoscopy is therefore a critical step in preventing the progression of CRC. Despite its importance, polyp detection during colonoscopies remains a difficult task for clinicians. The procedure is visually demanding, requiring continuous focus and careful inspection of every frame in a live video feed. Small, flat, or partially hidden polyps can be easily overlooked, especially under poor lighting, motion blur, or presence of bowel contents. In addition, detection accuracy often varies between practitioners due to differences in training and fatigue during long procedures, leading to missed or delayed diagnoses. To address these challenges, we propose a real-time computer-aided detection system based on the YOLOv8 algorithm. YOLOv8 was selected for its excellent balance between detection speed and accuracy, outperforming previous models like YOLOv5, Faster R-CNN, and SSD in demanding real- world settings. We further enhanced the system using transfer learning, allowing it to benefit from large-scale pre-trained datasets—an essential advantage in medical imaging, where labelled data is often scarce. Our model achieved a precision of 91.2%, recall of 89.7%, and an F1-score of 90.4%, marking a 12–15% performance increase over traditional methods. Additionally, we incorporated a feature to assess the probability of a polyp developing into cancer, adding valuable clinical insight beyond detection. By combining artificial intelligence with medical imaging, this system supports healthcare professionals in making faster, more accurate decisions ultimately making colorectal cancer screening more effective, consistent, and accessible.