Multiparametric dual-energy computed tomography radiomics for predicting microvascular invasion in hepatocellular carcinoma

Microvascular invasion (MVI) is a well-established predictor of poor prognosis in hepatocellular carcinoma (HCC), making its accurate preoperative diagnosis essential for optimizing treatment strategies. This study aimed to evaluate the potential of multiparametric dual-energy computed tomography (DECT) radiomics for the noninvasive prediction of MVI. Patients with pathologically confirmed primary HCC who underwent contrast-enhanced DECT were retrospectively enrolled. Radiomics features were extracted from virtual monochromatic images (VMI), iodine density (ID) maps, and effective atomic number (Zeff) maps for each phase, resulting in the VMI, ID, Zeff, and Combined MIZ (Monoenergetic, Iodine, Zeff) feature sets. In parallel, a total of 24 conventional quantitative parameters (e.g., iodine concentration and normalized iodine concentration) were measured on these parametric maps for benchmark comparison. Feature selection was performed using analysis of variance (ANOVA), minimum redundancy maximum relevance (mRMR), and least absolute shrinkage and selection operator (LASSO) for radiomics features, with univariate logistic regression for quantitative parameters. Predictive models were developed using random forest (RF), support vector machine (SVM), and extreme gradient boosting (XGBoost). Model performance was evaluated using receiver operating characteristic (ROC) analysis and the area under the curve (AUC), compared via the DeLong test. 126 patients (mean age, 56.79 ± 12.07 years; 113 men; 47 MVI-positive) were included. The radiomics model based on the Combined MIZ set achieved mean AUCs of 0.9129 in the training cohort and 0.8928 in the test cohort. Among the classifiers, XGBoost demonstrated the highest performance, with an AUC of 0.9427 (95% CI: 0.8995–0.9859) in the training cohort and 0.9375 (95% CI: 0.8681–1.000) in the test cohort. The Combined MIZ set demonstrated superior performance to that of the VMI, ID, Zeff, and quantitative parameter sets across all three classifiers (RF, SVM, and XGBoost), with all differences statistically significant (DeLong test, all p < 0.05). Multiparametric DECT radiomics shows promise in diagnosing MVI in HCC, demonstrating potential advantages over single-parametric radiomics and conventional quantitative parameters.