Development and Validation of a Computer-Aided Diagnostic System for Colorectal Lesions Based on the JNET...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Standardizing the Japan NBI Expert Team (JNET) classification for colorectal lesions is challenging due to endoscopist variability.
• A computer-aided diagnosis (CADx) system was developed to assist JNET diagnoses and facilitate the

Purpose of the Study:

• To evaluate the diagnostic performance of a CADx system for colorectal lesions.
• To compare the CADx system's performance against endoscopists with varying experience levels.
• To assess the CADx system's potential to support a

Main Methods:

• Two studies were conducted involving 43 patients (60 lesions) using CADx-assisted colonoscopies.
• Study 1 compared CADx and endoscopist JNET classification concordance with expert diagnosis.
• Study 2 evaluated CADx and endoscopist accuracy in identifying non-advanced neoplasia (<10 mm lesions) using histopathology as the gold standard.

Main Results:

• The CADx system achieved 83.3% concordance with expert diagnoses, outperforming beginners and intermediates.
• For lesions <10 mm, the CADx system's accuracy was 92.1%, significantly higher than all endoscopist groups.
• The CADx system demonstrated a 100% negative predictive value for advanced neoplasia.

Conclusions:

• The CADx system exhibits high diagnostic performance, comparable to expert endoscopists.
• The system's accuracy in diagnosing non-advanced neoplasia suggests its utility in supporting the