Effect of a Deep Learning-based Bile Duct Scanning System on the Diagnostic Accuracy of Common Bile Duct Stones During Examination by Novice Ultrasound Endoscopists

Overview

The bile duct scanning system based on deep learning can prompt endoscopists to scan standard stations and identify bile ducts and stones in real time. The purpose of this study is to evaluate the effectiveness and safety of the proposed deep learning-based bile duct scanning system in improving the diagnostic accuracy of common bile duct stones and reducing the rate of missed gallstones during bile duct scanning by novice ultrasound endoscopists in a single-center, tandem, randomized controlled trial

The incidence of gallstones has been increasing in recent years, up to 10-15% in developed countries, and is still increasing at a rate of 0.6% per year. It is estimated that common bile duct stones (CBDS) are present in about 10-20% of patients with symptomatic bile duct stones. Each year, common bile duct stones lead to acute complications such as biliary obstruction, cholangitis and acute pancreatitis in a large number of patients, seriously endangering their lives and health. In addition, Diagnosis Related Group (DRG) analysis shows that each episode of common bile duct stones costs $9,000, and acute pancreatitis that progresses from common bile duct stones can result in 275,000 hospitalizations annually, incurring $2.6 billion in costs and imposing a significant economic and health burden on society. Therefore, timely diagnosis of common bile duct stones and intervention for them is crucial. Endoscopic retrograde cholangiopancreatography (ERCP) is the method of choice for the diagnosis and treatment of CBDS, and guidelines recommend stone extraction for all patients with CBDS who are physically fit enough to tolerate ERCP operations. However, ERCP is a highly demanding and risky operation with the potential for serious complications such as PEP (incidence 2.6-3.5%). How to diagnose choledocholithiasis early and accurately, achieve timely intervention to improve prognosis, and avoid unnecessary medical operations to reduce risks are the challenges we are currently trying to solve. The guidelines recommend ultrasound endoscopy (EUS) or magnetic resonance cholangiopancreatography (MRCP) to determine the presence of CBDS, depending on the local level of care, for patients in the intermediate-risk group for CBDS and for patients in the low-risk group whose physicians still have a high suspicion of CBDS. sensitivity. In addition, a cost-effectiveness analysis showed that MRCP would be the preferred test when the predicted probability of CBDS is less than 40%, while EUS is the preferred test when the predicted probability is 40%-90%. Compared to MRCP, EUS has a wide range of applicability but a steep learning curve. ASGE states that a minimum of 225 EUS operations are required to qualify, while the ESGE states that a minimum of 300 operations are required. However, this experience can only be gained at training centers that perform a large number of cases. Thus, the training of novice physicians in resource-limited areas is a huge challenge, which leaves a significant shortage of experienced ultrasound endoscopists with poor performance in the actual diagnosis of common bile duct stones, greatly limiting the popularity of ultrasound endoscopy. The purpose of this study is to evaluate the effectiveness and safety of the proposed deep learning-based bile duct scanning system in improving the diagnostic accuracy of common bile duct stones and reducing the rate of missed gallstones during bile duct scanning by novice ultrasound endoscopists through a single-center, tandem, randomized controlled trial