Development and Multicenter Validation of a Contrast-Enhanced CT-Based Radiomics Model for Predicting Extrahepatic Metastasis after TACE in Unresectable Hepatocellular Carcinoma

Abstract Objective To develop and validate a prediction model integrating pretreatment contrast-enhanced CT (CECT) radiomics with clinical features, and to evaluate its value in predicting extrahepatic metastasis (EHM) after transarterial chemoembolization (TACE) in patients with unresectable hepatocellular carcinoma (HCC). Methods Clinical data of 152 patients with unresectable HCC who underwent TACE at the participating centers between January 2015 and December 2024 were retrospectively collected, and the patients were randomly assigned to a training set and an internal test set at a ratio of 7:3. Seventy additional HCC patients treated at other participating centers served as the external test set.Lesions were delineated on multiphase CECT images for radiomic feature extraction; after LASSO-based feature selection, the selected features were combined with independent clinical risk factors to construct prediction models and a nomogram. Model discrimination and calibration were assessed using receiver operating characteristic (ROC) curves, decision curve analysis (DCA), and calibration curves. The predictive performance of the models was validated using data from the external cohort. Results Multivariate analysis showed that age, alpha-fetoprotein (AFP), and maximum tumor diameter were independent predictors of EHM after TACE. Among the multiphase CT radiomics models, the venous-phase random forest (RF) model performed best (AUC = 0.899, 95% CI: 0.799–0.998), outperforming the logistic regression (LR) and support vector machine (SVM) models. The combined clinical-radiomics model achieved an AUC of 0.923 in the internal test set, outperforming the clinical model alone. In the multicenter external validation cohort, the combined model yielded an AUC of 0.878 (95% CI: 0.801–0.957), indicating good generalizability. Conclusion The clinical-radiomics combined model, constructed from selected CT radiomic features and clinical features, demonstrated robust performance in predicting EHM in both the single-center development cohort and the multicenter external validation cohort. It holds promise for estimating the probability of EHM in patients with HCC after TACE and provides a noninvasive tool for preoperative risk stratification.