Integrating multiphase MRI radiomics of tumor and peritumoral regions to predict early progression in Hepatocellular Carcinoma: A multicenter study.

OBJECTIVES

Hepatocellular carcinoma (HCC) remains a major cause of cancer-related mortality, with early progression (PD) after treatment posing a critical challenge. Reliable predictive models applicable to different treatment modalities are essential for risk stratification and individualized therapy.

MATERIALS AND

METHODS

In this multicenter retrospective study, 249 HCC patients from five institutions underwent curative resection or TACE with pre-treatment multiphase MRI. Radiomic features were extracted from intra- and peri-tumoral regions across arterial, portal venous, and subtraction delta phases. Feature selection was performed using mRMR and LASSO, followed by Cox regression to construct phase-specific and integrated multiphase radiomics model (MP-RM). Model performance was evaluated by C-index, time-dependent ROC, calibration, and decision curve analysis (DCA). SHAP values and visualization of specific radiomics features were applied to enhance interpretability.

RESULTS

MP-RM outperformed single-phase models in both training (1-year AUC 0.819; C-index 0.774) and external validation cohorts (1-year AUC 0.774; C-index 0.653). SHAP visualization identified key intra- and peri-tumoral radiomic features contributing to prediction. Risk stratification using MP-RM effectively separated high- and low-risk groups, with significantly shorter progression-free survival in high-risk patients (training HR = 3.95, p < 0.01; validation HR = 1.56, p = 0.05). Subgroup analyses supported relatively stable model performance across AFP levels, BCLC stages, treatment modalities, and age.

CONCLUSION

A multiphase MRI radiomics model integrating intra- and peri-tumoral features demonstrated moderate-to-good performance for predicting early progression in HCC patients treated with resection or TACE. The model may provide complementary information for risk stratification, although further prospective validation is required before clinical implementation.