Study on the Clinical Value of MRCP Imaging before ERCP for Uncertain Choledocholithiasis
Objective To investigate the clinical value of MRCP imaging before ERCP for uncertain choledocholithiasis.Methods The medical records of 120 patients with CBDS clinical manifestations or imaging examination(abdominal ultrasound or CT)considering the presence of CBDS in our hospital from February 2019 to April 2023 were collected.It is divided into training set and verification set.According to the results of ERCP,84 patients were divided into positive stone group(n=54)and negative stone group(n=30),and 36 cases were verified,including 25 cases in positive stone group and 11 cases in negative stone group.The confluence of pancreaticobiliary duct in MRCP was used as the region of interest,and the imaging features were extracted.After the collinearity was eliminated by Logistic regression,the clinical factors of CBDS were analyzed and the Clinic model was constructed;the optimal feature model(Rad model)was constructed by support vector machine(SVM);and the artificial neural network model(Combine model)was constructed by Python3.6 based on Softmax strategy.The discrimination,calibration and net benefit of each model of the training set and the verification set were evaluated by using the subject working characteristic curve(ROC),Hosmer-Lemeshow test and decision curve analysis.Results CRP(OR=10.306,95％CI:5.827～18.224,P＜0.001),cholesterol(OR=7.119,95％CI:3.066～12.935,P＜0.001),choledochal angle(OR=2.526,95％CI:1.430～7.284,P＜0.001),biliary tract infection(OR=3.064,95％CI:1.288～5.193,P＜0.05),choledochal dilatation(OR=5.289,95％CI:2.067～9.381,P＜0.05).All of them were independent influencing factors for predicting CBDS.The sensitivity and specificity of the Logistic clinical regression model were 72.83％and 67.52％,respectively.According to the 11 optimal features and their corresponding weighting coefficients,the imaging tag Radiomics score containing 11 optimal features was constructed by SVM model.There was significant difference in Radiomics score between positive stone group and negative stone group in training set(P＜0.001)and verification set(P＜0.037).The area under the curve of Combine model in training set and verification set was 0.962(95％CI:0.925～0.987)and 0.937(95％CI:0.851～0.993)respectively,which was significantly higher than that of Rad model and Clinic model in the same group by Delong test.Hosmer-Lemeshow test shows that the Combine model is consistent in the two datasets.Decision curve analysis shows that Combine model curve is significantly higher than Clinic model,Rad model and extreme curve.Conclusion CRP,cholesterol,choledochal angle ≤ 120°,biliary tract infection and choledochal dilatation are the clinical influencing factors of CBDS.The artificial neural network model constructed by MRCP imaging combined with clinical factors makes non-invasive prediction of ERCP in uncertain CBDS possible.Compared with traditional clinical diagnosis methods,it can enhance the diagnostic efficiency of CBDS prediction model and provide important clinical decision-making guidance.
NETL
NSTL
万方数据
维普
