Clinical Validation of a Generative AI System for Diagnosing Ampullary Lesions: A Multicenter Study

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate histologic classification of ampullary lesions is crucial for treatment.
• Conventional biopsy methods face limitations due to sampling errors and false negatives.
• Generative AI offers a potential solution to enhance diagnostic performance.

Purpose of the Study:

• To evaluate the clinical utility of a generative AI-based computer-aided diagnosis (CAD) system.
• To assess the AI system's ability to improve diagnostic accuracy using real and synthetic endoscopic images.
• To determine the impact of AI assistance on endoscopist performance.

Main Methods:

• A retrospective multicenter study used duodenoscopic images classified as Normal, Adenoma, or Cancer.
• A generative AI-based CAD system synthesized 500 images per class for data augmentation.
• A reader study compared diagnostic performance with and without AI assistance among expert and trainee endoscopists.

Main Results:

• The AI-CAD system achieved high overall diagnostic accuracy (91.57%) and adenoma identification accuracy (88.76%).
• CAD assistance significantly improved adenoma sensitivity (63.47% to 70.56%) and predictive values.
• AI support reduced interobserver variability and improved classification consistency for all lesion types.

Conclusions:

• Generative AI-based CAD systems enhance diagnostic accuracy and consistency for ampullary lesions.
• The AI system shows potential as a valuable adjunct to routine duodenoscopy.
• Improved adenoma recognition and therapeutic decision-making are key benefits of this AI tool.