Real-time detection of colon polyps during colonoscopy using deep learning: systematic validation with four...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Colorectal cancer (CRC) is a significant global health concern.
• Early polyp detection during colonoscopy is crucial for CRC prevention.
• Manual polyp detection can be challenging and prone to missed lesions.

Purpose of the Study:

• To develop and validate a deep learning algorithm for automatic polyp detection.
• To assess the algorithm's performance in terms of sensitivity and false positive rates.
• To evaluate the algorithm's potential to assist endoscopists in clinical practice.

Main Methods:

• A YOLOv2 deep learning model was trained on 8,075 images containing 503 polyps.
• Algorithm validation was performed on three distinct datasets (A, B, C) including static images and videos.
• Median filtering was applied to video analysis to reduce false positives.

Main Results:

• Per-image polyp detection sensitivity reached 96.7% (Dataset A) and 90.2% (Dataset B).
• Video analysis sensitivity was 87.7% (Dataset C), with a false positive rate reduced to 6.3% using median filtering.
• The algorithm identified all polyps found by endoscopists and 7 additional polyps, operating at 67.16 frames per second.

Conclusions:

• The developed deep learning algorithm demonstrates high sensitivity and rapid processing for automatic polyp detection.
• The algorithm shows promise as a supportive tool for endoscopists, potentially enhancing polyp detection rates.
• Further integration into clinical workflows could improve colonoscopy outcomes and CRC prevention.