CT-Based Habitat Model for Predicting Tumor Response and Survival in Hepatocellular Carcinoma Treated with Transarterial Chemoembolization Combining Molecular Targeted Agents and Immune Checkpoint Inhibitors.

Abstract

Rationale and objectives: To develop and validate a CT-based habitat model for predicting tumor response and overall survival (OS) in patients with unresectable hepatocellular carcinoma (uHCC) undergoing transarterial chemoembolization, combined with molecular-targeted agents, and immune checkpoint inhibitors (TACE-MTAs-ICIs).

Materials and methods: A total of 200 patients treated with TACE-MTAs-ICIs between June 2019 and August 2024 were retrospectively included. Voxel-level radiomic features were extracted from contrast-enhanced CT images, and tumor habitats were identified using K-means clustering. Radiomic features were extracted from both habitat subregions and the entire tumor volume (conventional radiomics). A support vector machine (SVM) model was developed to predict tumor response, with SHapley Additive exPlanations (SHAP) analysis applied for interpretability. In parallel, a Cox proportional hazards model was constructed to predict OS. Independent clinical risk factors were incorporated with radiomic features to build a combined model. Model performance was evaluated and compared using multiple statistical metrics.

Results: In the test cohort, the habitat model achieved strong performance for tumor response prediction (AUC: 0.881) and OS stratification (C-index: 0.703; 1-year AUC: 0.788; 2-year AUC: 0.806), outperforming the conventional radiomics model. Notably, the integrated model combining habitat features and clinical variables further improved predictive accuracy, yielding an AUC of 0.884 for response prediction and superior OS discrimination (C-index: 0.749; 1-year AUC: 0.831; 2-year AUC: 0.841).

Conclusion: The proposed CT-based habitat model enables a more accurate and interpretable assessment of treatment response and OS in HCC, offering valuable noninvasive biomarkers that reflect intra-tumor heterogeneity. This approach holds promise for improving individualized treatment planning and clinical outcomes.

Data availability statement: The data that support the findings of this study are available from the corresponding author upon reasonable request.