Purpose: To explore whether texture features based on magnetic resonance can distinguish diseases combined hepatocellular-cholangiocarcinoma (cHCC-CC) from hepatocellular carcinoma (HCC) before operation.
Methods: The clinical baseline data and MRI information of 342 patients with pathologically diagnosed cHCC-CC and HCC in two medical centers were collected. The data were divided into the training set and the test set at a ratio of 7:3. MRI images of tumors were segmented with ITK-SNAP software, and python open-source platform was used for texture analysis. Logistic regression as the base model, mutual information (MI) and Least Absolute Shrinkage and Selection Operator (LASSO) regression were used to select the most favorable features. The clinical, radiomics, and clinic-radiomics model were constructed based on logistic regression. The model's effectiveness was comprehensively evaluated by the receiver operating characteristic (ROC) curve, area under the curve (AUC), sensitivity, specificity, and Youden index which is the main, and the model results were exported by SHapley Additive exPlanations (SHAP).
Results: A total of 23 features were included. Among all models, the arterial phase-based clinic-radiomics model showed the best performance in differentiating cHCC-CC from HCC before an operation, with the AUC of the test set being 0.863 (95% CI: 0.782 to 0.923), the specificity and sensitivity being 0.918 (95% CI: 0.819 to 0.973) and 0.738 (95% CI: 0.580 to 0.861), respectively. SHAP value results showed that the RMS was the most important feature affecting the model.
Conclusion: Clinic-radiomics model based on DCE-MRI may be useful to distinguish cHCC-CC from HCC in a preoperative setting, especially in the arterial phase, and RMS has the greatest impact.
Keywords: differential diagnosis; machine learning; magnetic resonance; primary liver cancer; radiomics.
© 2023 Guo et al.