Multiphase MRI-Derived Delta radiomics integrated with clinical features for survival prediction in hepatocellular carcinoma after thermal ablation

Abstract

Rationale and objectives

To evaluate the predictive value of a delta radiomic score (DRS)derived from multiphase contrast-enhanced Magnetic Resonance Imaging (MRI), in combination with clinical characteristics, forecasting overall survival (OS) and recurrence-free survival (RFS) in patients with hepatocellular carcinoma (HCC) after thermal ablation.

Materials and Methods

This retrospective study enrolled 415 patients with HCC from two centers who underwent multiphase contrast-enhanced magnetic resonance imaging and percutaneous thermal ablation. Patients in Institution 1 (n = 315) were randomly assigned to the training cohorts (n = 220) and internal tests (n = 95), while 100 patients in Institution 2 formed the external validation cohort. Delta radiomic features were extracted by comparing contrast-enhanced phases with noncontract images, followed by variance filtering, correlation analysis, random forest ranking, and Cox regression to derive the DRS. The DRS was combined with clinical characteristics, including tumor burden score (TBS), Barcelona Clinic Liver Cancer (BCLC), and China Liver Cancer Staging (CNLC), to build Random Survival Forest (RSF) models. The performance of the model was evaluated using the concordance index (C-index), the area under the curve (AUC), the calibration curves and the analysis of decision curves (DCA).

Results

The RSF models that integrate DRS with clinical characteristics demonstrated favorable predictive performance for both OS and RFS. For OS predictions at 36 and 60 months, AUCs ranged from 0.71 to 0.83 in cohorts, with corresponding C indices between 0.69 and 0.83. For the prediction of short-term RFS at 6, 12, and 24 months, the AUC ranged from 0.71 to 0.91, with C-indices between 0.68 and 0.71. Calibration and DCA analysis confirmed the robustness and clinical utility of the models in internal and external validations.

Conclusions

Delta radiomics derived from multiphase MRI effectively captured tumor dynamics and heterogeneity. When integrated with clinical characteristics, the resulting models allowed an accurate prediction of both recurrence and long-term survival, providing a practical tool for risk stratification and individualized post-ablation management in HCC.