IVIM-DWI-based radiomic model for preoperative prediction of hepatocellular carcinoma differentiation

Objective To construct a radiomic model based on intravoxel incoherent motion diffusion-weighted imaging (IVIM-DWI) for preoperative prediction of hepatocellular carcinoma (HCC) differentiation and validate its clinical value. Methods Clinical and imaging data of 187 HCC patients who received surgical treatment in the Third Affiliated Hospital of Chongqing Medical University from June 2018 to January 2024 were collected and retrospectively analyzed. According to the postoperative pathological results, they were divided into low-differentiation group (n=58) and non-low-differentiation group (n=129), and randomly divided into the training group (n=149) and the validation group (n=38) with the ratio of 8 ∶2. Univariate analysis was used to assess the clinical indicators related to HCC differentiation, and then a clinical model was constructed. Pyramidimics software was used to extract the radiomic features of IVIM-DWI functional images, and minimum absolute contraction and selection operator logistic regression algorithm were employed to screen those highly correlated indicators with HCC differentiation. Support vector machine (SVM), logistic regression (LR) and random forest (RF) algorithms were utilized to construct different image omics models. SVM algorithm was applied to construct the combined imaging omics and clinical model. The internal verification of the model was carried out by using ten-fold cross-validation. Receiver operating characteristic (ROC) curve, area under the curve (AUC), calibration curve, and decision curve analysis (DCA) were used to evaluate the diagnostic value and clinical benefits of clinical model, radiomic model, and their combination. Results A total of 4 060 radiological features were extracted, and after feature screening and dimensionality reduction, 24 features were finally included to construct the model. Among all models, the predictive performance of the radiomic model and the radiomic-clinical combined model was better than that of the clinical model. In the comparison between the radiomic model constructed by SVM algorithm and the radiomics-clinical combined model, the AUC value was 0.954 (0.908~1.000) for the former model, and was 0.943 (0.905~0.982) for the latter model in the training set, and there was no significant difference between them. In the validation set, the AUC value was 0.807 (0.640~0.975) and 0.876 (0.743~1.000), respectively, with statistical difference between the 2 models (P < 0.05). Calibration curve analysis showed that the radiomic model and the radiomics-clinical combined model had good goodness of fit. DCA indicated that the net benefit was higher and the threshold probability range was larger in both the radiomic model and the radiomics-clinical combined model, and the net benefit of the combined model was larger. Conclusion Both the radiomic model and the combined radiomics-clinical model constructed on the basis of IVIM-DWI functional images can better predict the severity of HCC differentiation before surgery.