MRI-based prediction of very early recurrence within 1 year after resection of HCC: An international cohort study

BACKGROUND & AIMS: Very early recurrence (within 1 year after resection) of hepatocellular carcinoma (HCC) indicates aggressive tumor biology and poor prognosis; however, non-invasive prediction tools remain scarce. This study aimed to develop and validate MRI-based models for predicting very early recurrence and to explore their biological underpinnings. METHODS: This international, multicenter cohort study retrospectively included 1,811 consecutive patients who underwent curative-intent resection for a single BCLC 0/A HCC and preoperative contrast-enhanced MRI or CT at 14 tertiary care hospitals in China, South Korea, Singapore, France, and the USA, as well as 40 eligible patients from the TCGA-LIHC dataset. To predict 1-year recurrence-free survival (RFS), MRI for Very Early Recurrence Prediction (MERP) models were developed using either preoperative variables alone (MERP-pre) or a combination of pre- and postoperative variables (MERP-post) in the derivation cohort (n = 628). The models were externally validated in Eastern (n = 775) and Western (n = 178) test cohorts and further evaluated in a CT test cohort (n = 230). Biological correlates were investigated using whole-exome and RNA sequencing. RESULTS: MERP-pre incorporated alpha-fetoprotein, tumor size, portal venous phase peritumoral hypoenhancement, and intratumoral blood products, whereas MERP-post replaced tumor size with microvascular invasion (MVI). Both models outperformed major staging systems and the IMbrave050 criteria across all test cohorts (C-index, 0.685-0.790 vs. 0.524-0.682; p values, <0.001 to 0.046) and improved risk reclassification (net reclassification improvement [NRI], 0.004-0.372). The MERP models consistently stratified patients into high- and low-risk groups with distinct 1-year RFS rates (all p <0.001). MERP-pre further improved MVI-based risk reclassification (NRI, 0.323) and survival risk stratification. Tumors classified as low risk by MERP-pre exhibited features of the non-proliferative class, increased lipid catabolism, and an immunoactive microenvironment. CONCLUSIONS: The MERP models may serve as robust, generalizable, and biologically informed tools for predicting very early HCC recurrence and may help refine risk-stratified management strategies. IMPACT AND IMPLICATIONS: Very early recurrence (VER) of hepatocellular carcinoma (HCC) within 1 year after resection reflects aggressive tumor biology and is associated with poor prognosis, yet reliable tools for predicting VER remain scarce. In this large, international, multicenter cohort of patients with a single Barcelona Clinic Liver Cancer stage 0/A HCC undergoing curative-intent resection, we developed and externally validated two MRI-based models (MERP-pre and MERP-post) that integrate routinely available clinicopathological and imaging variables. The MERP models consistently outperformed existing prognostic systems and effectively stratified patients into high- and low-risk groups with distinct 1-year recurrence-free survival across diverse cohorts. These findings highlight the potential of the MERP models to complement existing prognostic systems and optimize risk-stratified management of HCC.