Beyond Human Vision: Revolutionizing the Localization of Diminutive Sessile Polyps in Colonoscopy

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Colorectal cancer (CRC) is a significant global health concern with increasing incidence.
• Early detection and removal of polyps, precursors to CRC, are vital for prevention.
• Traditional colonoscopy is effective but susceptible to human error in polyp detection.

Purpose of the Study:

• To develop and evaluate an AI-based system for localizing diminutive sessile polyps in colonoscopy images.
• To enhance the accuracy of polyp detection, thereby aiding in colorectal cancer prevention.

Main Methods:

• Utilized the YOLO-V8 deep learning model for polyp localization.
• Assembled a diverse dataset from multiple sources, including white light endoscopy (WLE) and narrow-band imaging (NBI) images.
• Performed comprehensive evaluations and analyzed dataset suitability using polyp size and coordinate matrices.

Main Results:

• Achieved high performance metrics: 96.4% precision, 93.89% recall, and 94.46% F1-score.
• Demonstrated the effectiveness of a balanced hyperparameter combination and a comprehensive dataset.
• Validated the dataset's suitability for colorectal polyp localization.

Conclusions:

• The AI-based polyp localization assistant shows significant promise for improving the detection of diminutive sessile colorectal polyps.
• This technology can advance colorectal cancer diagnosis, particularly in offline settings and for analyzing endoscopy capsule images.
• The study highlights the potential of AI to augment gastroenterologists' diagnostic capabilities and reduce errors.