Cholangioskopie im Fokus: Künstliche Intelligenz verbessert Genauigkeit und Effizienz der Diagnoseverfahren

Background: We aimed to develop a convolutional neural network (CNN) model for detecting neoplastic lesions during real-time digital single-operator cholangioscopy (DSOC) and to clinically validate the model through comparisons with DSOC expert and nonexpert endoscopists. Methods: In this two-stage study, we first developed and validated CNN1. Then, we performed a multicenter diagnostic trial to compare four DSOC experts and nonexperts against an improved model (CNN2). Lesions were classified into neoplastic and non-neoplastic in accordance with Carlos Robles-Medranda (CRM) and Mendoza disaggregated criteria. The final diagnosis of neoplasia was based on histopathology and 12-month follow-up outcomes. Results: In stage I, CNN2 achieved a mean average precision of 0.88, an intersection over the union value of 83.24 %, and a total loss of 0.0975. For clinical validation, a total of 170 videos from newly included patients were analyzed with the CNN2. Half of cases (50 %) had neoplastic lesions. This model achieved significant accuracy values for neoplastic diagnosis, with a 90.5 % sensitivity, 68.2 % specificity, and 74.0 % and 87.8 % positive and negative predictive values, respectively. The CNN2 model outperformed nonexpert #2 (area under the receiver operating characteristic curve [AUC]-CRM 0.657 vs. AUC-CNN2 0.794, P < 0.05; AUC-Mendoza 0.582 vs. AUC-CNN2 0.794, P < 0.05), nonexpert #4 (AUC-CRM 0.683 vs. AUC-CNN2 0.791, P < 0.05), and expert #4 (AUC-CRM 0.755 vs. AUC-CNN2 0.848, P < 0.05; AUC-Mendoza 0.753 vs. AUC-CNN2 0.848, P < 0.05).