LAUSR

BACKGROUND AND AIMS Diagnosis and characterization of biliary strictures is challenging. The introduction of digital single-operator cholangioscopy (DSOC) allowing direct visual inspection of the lesion and targeted biopsies significantly improved the diagnostic yield in patients with indeterminate biliary strictures. However, the diagnostic efficiency of DSOC remains suboptimal. Convolutional neural networks (CNNs) have shown great potential for the interpretation of medical images. We aimed to develop a CNN-based system for automatic detection of malignant biliary strictures in DSOC images. METHODS We developed, trained, and validated a CNN based on DSOC images. Each frame was labeled as normal/benign findings or as a malignant lesion if histopathological evidence of biliary malignancy was available. The entire dataset was split for 5-fold cross validation. In addition, the image dataset was split for constitution of training and validation datasets. The performance of the CNN was measured by calculating the area under the receiving operating characteristic curve (AUC), sensitivity, specificity, positive, and negative predictive values (PPV and NPV, respectively). RESULTS A total of 11855 images from 85 patients were included (9695 of malignant strictures and 2160 of benign findings). The model had an overall accuracy of 94.9%, a sensitivity of 94.7%, a specificity of 92.1% and an AUC of 0.988 in cross-validation analysis. The image processing speed of the CNN was 7 ms/frame. CONCLUSIONS The developed deep learning algorithm accurately detected and differentiated malignant strictures from benign biliary conditions. The introduction of artificial intelligence algorithms to DSOC systems may significantly increase its diagnostic yield for malignant strictures.