Diagnostic performance of real-time artificial intelligence using deep learning analysis of endoscopic ultrasound videos for gallbladder polypoid lesions

This study developed and evaluated a deep learning-based real-time artificial intelligence (AI) for the diagnosis of gallbladder polypoid lesions. An AI model was constructed to distinguish between neoplastic and non-neoplastic polyps by analyzing endoscopic ultrasound (EUS) videos. The research team used a total of 1,234 EUS images for training, and the AI demonstrated high diagnostic accuracy. Specifically, the AI achieved a sensitivity of 92.3% and a specificity of 89.7%. The Aview system (Coreline Soft) enabled real-time analysis of EUS videos, allowing clinicians to obtain diagnostic information immediately. This AI technology can help reduce unnecessary surgeries for gallbladder polyps and assist in making appropriate treatment decisions. The ground truth for segmentation was manually annotated by four EUS experts who had performed more than 500 EUS procedures in pancreatobiliary imaging, using Aview software (Coreline Soft, Seoul, Republic of Korea). These findings highlight the potential of real-time AI in endoscopic imaging, supporting immediate decision-making and improving workflow efficiency in the evaluation of gallbladder lesions. In this study, Aview software was utilized for annotation and real-time analysis of EUS data, supporting consistent interpretation of ultrasound-based imaging.