Predictive Value Of Preoperative MRCP Based Clinical-radiomics Model For Post-ERCP Pancreatitis

Objective:Endoscopic retrograde cholangiopancreatography(ERCP)has been widely applied in the diagnosis and minimally invasive management of pancreatic and biliary diseases.Post-ERCP pancreatitis(PEP)is the most frequent ERCP-related adverse event.Some patient-or procedure-related risk factors have been used to establish predictive models for PEP.However,medical imaging data was rarely used.The primary aim was to develop and validate a clinical-radiomics model based on preoperative magnetic resonance cholangiopancreatography(MRCP)for the prediction of PEP.Methods:A retrospective review of patients who underwent ERCP at a tertiary care center was performed.Preoperative MRCP and clinical risk factors for PEP were collected.Pancreaticobiliary duct junctions were outlined manually in MRCP.Radiomics features were extracted according to the imaging biomarker standardization initiative.The least absolute shrinkage and selection operator regression were performed for feature reduction.Binary logistic regression was used to develop clinical models,radiomics models,and clinical-radiomics models for PEP prediction pre-and post-ERCP.The performance of different models was compared.Five-fold cross-validations and bootstrap resampling analysis were applied for validation of the model with the best comprehensive performance.The final model was converted into a nomogram.Results:A total of 192 patients(67 PEP patients,125 non-PEP patients)were randomized into the training set(133 patients)and the validation set(59 patients).For the pre-ERCP models,the radiomics model yielded a higher area under the curve(AUC)of 0.707(95% confidence interval [CI],0.561-0.853)in the validation set than the clinical model.However,the difference was not statistically significant.For the post-ERCP models,the clinical-radiomics model achieved a significantly higher AUC of 0.726(95% CI,0.581-0.870)in the validation set than the clinical model(0.505,95% CI,0.364-0.646).The final model,including five clinical risk factors and seven radiomics features,yielded an optimism-adjusted AUC of 0.778.Conclusion:The MRCP radiomics features of the pancreaticobiliary junction have predictive value for PEP.The combination of MRCP radiomics features and clinical risk factors could improve the model performance.Hence,this model provides an alternative for the early prediction of PEP.