Triphasic CT Radiomics Model for Preoperative Prediction of Hepatocellular Carcinoma Pathological Grading

This study aimed to develop and validate a triphasic CT-based radiomics model for the synchronous prediction of multiple critical pathological markers in hepatocellular carcinoma (HCC).

This retrospective study analyzed 174 patients with 187 hepatocellular carcinoma (HCC) lesions. Radiomic features (n = 2264) were extracted from arterial phase (AP), venous phase (VP), and delayed phase (DP) CT images. Key features were selected using minimum redundancy maximum relevance (mRMR), SelectKBest, and least absolute shrinkage and selection operator (LASSO) algorithms. Logistic regression and support vector machine (SVM) classifiers were employed to develop individual phase-specific models and a triphasic fusion model. Model performance was evaluated through the area under the curve (AUC), sensitivity, specificity, decision curve analysis, and other metrics.

The triphasic fusion model demonstrated superior performance. In the testing 1 dataset, the triphasic fusion model achieved AUCs of 0.890 (95% CI: 0.741-1), 0.895 (95% CI: 0.781-1) and 0.829 (95% CI: 0.675-0.984) for Edmondson-Steiner (Ed) grading, Microvascular invasion (MVI) grading, and Satellite nodule (SN) grading, respectively. In the testing 2 (validation) dataset, the triphasic fusion model achieved AUCs of 0.836 (95% CI: 0.739-0.934), 0.871 (95% CI: 0.748-0.993) and 0.810 (95% CI: 0.656-0.963) for Ed, MVI, and SN grading, respectively. The performance of the fusion model was better than that of the single-phase models.

The triphasic CT radiomics model provides a noninvasive tool for preoperative prediction of HCC pathological grading (Ed, MVI, SN), enhancing diagnostic accuracy for clinical decision-making and prognostic evaluation.