The value of CT-based radiomics in predicting the prognosis of acute pancreatitis

Early judgment of the progress of acute pancreatitis (AP) and timely intervention are crucial to the prognosis of patients. The purpose of this study was to investigate the application value of CT-based radiomics of pancreatic parenchyma in predicting the prognosis of early AP. This retrospective study enrolled 137 patients diagnosed with AP (95 cases in the progressive group and 42 cases in the non-progressive group) who underwent CT scans. Patients were randomly divided into a training set ( = 95) and a validation set ( = 42) in a ratio of 7: 3. The region of interest (ROI) was outlined along the inner edge of the pancreatic parenchyma manually, and the Modified CT Severity Index (MCTSI) was assessed. After resampling and normalizing the CT image, a total of 2,264 radiomics features were extracted from the ROI. The radiomics features were downscaled and filtered using minimum redundancy maximum correlation (mRMR) and the least absolute shrinkage and selection operator algorithm (LASSO) regression, in turn, and the more optimal subset of radiomics features was selected. In addition, the radiomics score (rad-score) was calculated for each patient by the LASSO method. Clinical data were also analyzed to predict the prognosis of AP. Three prediction models, including clinical model, radiomics model, and combined clinical-radiomics model, are constructed. The effectiveness of each model was evaluated using receiver operating characteristic (ROC) curve analysis. The DeLong test was employed to compare the differences between the ROC curves. The decision curve analysis (DCA) is used to assess the net benefit of the model. The mRMR algorithm and LASSO regression were used to select 13 radiomics features with high values. The rad-score of each texture feature was calculated to fuse MCTSI to establish the radiomics model, and both the clinical model and clinical-radiomics model were established. The clinical-radiomics model showed the best performance, the AUC and 95% confidence interval, accuracy, sensitivity, and specificity of the clinical-radiomics model in the training set were 0.984 (0.964-1.000), 0.947, 0.955, and 0.931, respectively. In the validation set, they were 0.942 (0.870-1.000), 0.929, 0.966, and 0.846, respectively. The Delong test showed that the predictive efficacy of the clinical-radiomics model was higher than that of the clinical model ( = 2.767, = 0.005) and the radiomics model ( = 2.033, = 0.042) in the validation set. Decision curve