Automatic detection of tumor vessels in indeterminate biliary strictures in digital single-operator cholangioscopy

Abstract

Background and study aims Indeterminate biliary strictures pose a significative clinical challenge. Dilated, irregular, and tortuous vessels, often described as tumor vessels, are frequently reported in biliary strictures with high malignancy potential during digital single-operator cholangioscopy (D-SOC). In recent years, the development of artificial intelligence (AI) algorithms for application to endoscopic practice has been intensely studied. We aimed to develop an AI algorithm for automatic detection of tumor vessels (TVs) in D-SOC images. Patients and methods A convolutional neural network (CNN) was developed. A total of 6475 images from 85 patients who underwent D-SOC (Spyglass, Boston Scientific, Marlborough, Massachusetts, United States) were included. Each frame was evaluated for the presence of TVs. The performance of the CNN was measured by calculating the area under the curve (AUC), sensitivity, specificity, positive and negative predictive values. Results The sensitivity, specificity, positive predictive value, and negative predictive value were 99.3 %, 99.4 %, 99.6% and 98.7 %, respectively. The AUC was 1.00. Conclusions Our CNN was able to detect TVs with high accuracy. Development of AI algorithms may enhance the detection of macroscopic characteristics associated with high probability of biliary malignancy, thus optimizing the diagnostic workup of patients with indeterminate biliary strictures.

Artificial intelligence for diagnosing neoplasia on digital cholangioscopy: development and multicenter validation of a convolutional neural network model

“…The authors extracted frames from these patients in both studies (11855 and 6475, respectively) and divided the frames into training (80% of the frames) and testing datasets (the remaining 20%). In their studies, they obtained the diagnostic accuracy of their models in terms of frames, with 94.7% sensitivity, 92.1% specificity, 94.8% positive predictive value, and 84.2% negative predictive value [16], with similar results in the second study [15]. However, these models could not be applied in prerecorded videos nor live procedures.…”

“…To date, despite the numerous advantages of DSOC, there is an ongoing discrepancy between the operators' visual impression using current classifications for indeterminate biliary lesions. To overcome this limitation, the application of new technologies to aid in image interpretation has been proposed; however, the proposed models could only be applied in images [15,16]. In the present study, we developed a new DSOC-based CNN for recognizing neoplasia in indeterminate biliary lesions in prerecorded videos and real-time DSOC procedures and compared the model with DSOC experts and nonexperts using the CRM and Mendoza classifications.…”

“…These results are similar to those of two recent studies by Pereira et al and Saraiva et al, who presented ▶ Table 3 Comparison of diagnostic accuracy between experts using the Carlos Robles-Medranda and Mendoza classifications and convolutional neural network model version 2 in clinical validation. CNN models that detected tumor vessels during DSOC using 85 subjects [15,16]. The authors extracted frames from these patients in both studies (11 855 and 6475, respectively) and divided the frames into training (80 % of the frames) and testing (the remaining 20 %) datasets.…”

“…To date, despite the numerous advantages of DSOC, there is an ongoing discrepancy between operators' visual impressions using current classifications for indeterminate biliary lesions. To overcome this limitation, the application of new technologies to aid image interpretation has been proposed; however, the proposed models could only be applied to images [15,16].…”

A Comparative Analysis of Artificial Intelligence-based Digital Cholangioscopy and Probe-based Confocal Laser Endomicroscopy for Detecting Malignant Bile Duct Lesions: A single-centre retrospective study

“…Diagnostic accuracies comparison were defined through DeLong’s test for two ROC curves. 1 2 3 4 5 6 7 8 9 10 11 .…”