S50 Decoding Biliary Strictures: A Transatlantic Multicenter Study on AI Performance Across Lesion Topography

Introduction: Cholangiocarcinoma (CCa) is a complex neoplasm with poor prognosis, being typically divided based on location in intrahepatic (iCCa), perihilar (pCCa) or distal (dCCa). Digital single-operator cholangioscopy (DSOC) is fundamental in CCa diagnosis despite some known limitations, such as suboptimal biopsy sensitivity, lack of a standardized morphology classification and technical issues. Emerging evidence highlights AI’s potential to enhance DSOC diagnostic capacity, but the differential performance according to tumor characteristics like topography remains unassessed. This study aimed to evaluate an AI algorithm’s effectiveness in detecting iCCa, pCCa, and dCCa. Methods: A deep learning model based on a convolutional neural network was designed to predict the malignancy status of biliary strictures. The model was built using DSOC images from 6 international high-volume centers in 4 countries. Images showing strictures due to histologically proven cholangiocarcinoma or benign conditions were included. The detection rate of the algorithm was compared for each location category. Results: A total of 201,016 images from 72 patients and 2 different devices (SpyGlass™ DS II Direct Visualization System, EyeMax™ Cholangioscope) were included. Forty-five patients were men (62.5%) and the mean age was 67 years (SD: 12.6 years). Common bile duct strictures represented 51% of the total while intrahepatic and perihilar accounted for 25% and 24%, respectively. The algorithm’s detection rate of strictures representing CCa was 93.7% for iCCa, 94.0% for pCCa and 95.3% for dCCa. Pairwise comparisons revealed a significant difference in the detection rate between intrahepatic and perihilar (P < 0.001) and perihilar and common bile duct (P < 0.001). Conclusion: This is the first model to demonstrate accuracy across different types of biliary strictures and in diverse demographic populations, establishing a benchmark for AI-driven diagnostic tools in this domain. Topographic traits of biliary strictures influence the diagnostic performance of DSOC, often with difficult evaluation of intrahepatic and perihilar biliary strictures. While AI shows promise in improving the diagnostic capacity of DSOC, its performance is impacted by DSOC technical limitations. Developing models should address these issues before widespread clinical integration. This evidence stresses the importance of keeping a “human on the loop,” providing clinical reasoning to guide the application of AI algorithms.