The prediction of common bile duct stones using a neural network

Background: The role of preoperative ERCP and endoscopic sphincterotomy (ES) in the diagnosis and treatment of suspected common bile duct stones (CBDS) in the laparoscopic age is controversial. The preoperative diagnosis of CBDS by ERCP and the removal of CBDS by ES are advantageous because of technical difficulties in performing laparoscopic exploration of the common bile duct. Approximately 50% of preoperative ERCP examinations are normal, however. The noninvasive diagnosis of CBDS has assumed new importance, but it has proved to be an elusive goal. Neural networks are a form of artificial computer intelligence that have been used successfully to interpret ECGs and to diagnose myocardial infarcts. The purpose of this study was to determine whether a neural network could be trained to predict CBDS accurately in patients at high risk of having duct stones.

Study Design: We trained a back-propagation neural network to predict the presence of CBDS. Retrospective data from patients who had a cholecystectomy and either a preoperative ERCP or intraoperative cholangiogram were used to build the network, and it was tested using unseen data.

Results: One hundred forty patients were used to train the network, and 16 patients were used to test it. The trained network was able to predict CBDS in 100% of the patients in both the training and test sets.

Conclusions: Screening of high-risk patients for CBDS by neural network analysis is highly accurate. This promising new, noninvasive, and inexpensive technique can potentially decrease the need for preoperative ERCP by 50%, but additional prospective evaluation is indicated.