Development and validation of an artificial intelligence-based system for predicting colorectal cancer invasion depth...

Area of Science:

• Gastroenterology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate prediction of colorectal cancer invasion depth is crucial for effective treatment of large sessile polyps.
• Endoscopic optical prediction is often limited by endoscopist expertise and inter-observer variability.

Purpose of the Study:

• To develop a clinically applicable artificial intelligence (AI) system for identifying cancer invasion in large sessile colorectal polyps.

Main Methods:

• A deep learning-based colorectal cancer invasion calculation (CCIC) system was developed using multi-modal data (clinical information, white light, and image-enhanced endoscopy).
• The system was trained on 339 lesions and validated on 198 lesions across three hospitals, with performance evaluated through man-machine contests, reader studies, and video validation.

Main Results:

• The CCIC system achieved an overall accuracy of 90.4% (image) and 89.7% (video).
• CCIC performance was comparable to expert endoscopists and superior to senior and junior endoscopists.
• Augmentation with CCIC significantly improved junior endoscopists' accuracy from 75.4% to 85.3% (P = 0.002).

Conclusions:

• The deep learning-based CCIC system shows significant potential in predicting colorectal cancer invasion depth.
• This AI system can aid in determining optimal treatment strategies for large sessile colorectal polyps.