Preoperative MRI Intratumoral Heterogeneity Radiomics for Microvascular Invasion and Recurrence-Free Survival in HCC.
Microvascular invasion (MVI) is a major determinant of postoperative recurrence and poor prognosis in hepatocellular carcinoma (HCC), yet MVI cannot be reliably identified before surgery: pathologic evaluation is invasive, prone to sampling bias, and available only after resection. We aimed to develop and evaluate a preoperative radiomics biomarker for noninvasive MVI prediction and postoperative risk stratification of recurrence-free survival.
In this retrospective multicenter study, 567 patients with HCC from four centers were included. Based on pretreatment MRI and clinical variables available before surgery, radiomics models characterising the global tumour region (GTR) and intratumoral heterogeneity (ITH) were developed and integrated into a Fusion model for preoperative prediction of MVI. Discrimination, calibration, and clinical utility were assessed using AUC, calibration curves, and decision-curve analysis (DCA). Prognostic stratification was evaluated using C-index, time-dependent ROC analysis, and Kaplan-Meier analysis of recurrence-free survival, and benchmarked against the Barcelona Clinic Liver Cancer (BCLC), China Liver Cancer (CNLC), and American Joint Committee on Cancer (AJCC) staging systems.
The Fusion model showed higher discrimination than comparators in the validation and external test sets, achieving AUCs of 0.924 and 0.895, respectively. Ablation analyses showed incremental gains with the sequential addition of GTR features, ITH features, and clinical covariates. Fusion score-based stratification identified high- and low-risk groups with markedly divergent recurrence-free survival. In the external test set, the prognostic model achieved a C-index of 0.766 and time-dependent AUCs of 0.825 at 2 years and 0.878 at 5 years, showing higher discrimination than BCLC, CNLC, and AJCC staging (all p < 0.001). Performance remained robust across HBV status, histologic differentiation, and BCLC stage.
A radiomics biomarker integrating global tumour phenotype, intratumoral heterogeneity, and preoperative clinical variables enables noninvasive preoperative MVI assessment and may complement morphology-based staging for recurrence risk stratification in HCC.
In this retrospective multicenter study, 567 patients with HCC from four centers were included. Based on pretreatment MRI and clinical variables available before surgery, radiomics models characterising the global tumour region (GTR) and intratumoral heterogeneity (ITH) were developed and integrated into a Fusion model for preoperative prediction of MVI. Discrimination, calibration, and clinical utility were assessed using AUC, calibration curves, and decision-curve analysis (DCA). Prognostic stratification was evaluated using C-index, time-dependent ROC analysis, and Kaplan-Meier analysis of recurrence-free survival, and benchmarked against the Barcelona Clinic Liver Cancer (BCLC), China Liver Cancer (CNLC), and American Joint Committee on Cancer (AJCC) staging systems.
The Fusion model showed higher discrimination than comparators in the validation and external test sets, achieving AUCs of 0.924 and 0.895, respectively. Ablation analyses showed incremental gains with the sequential addition of GTR features, ITH features, and clinical covariates. Fusion score-based stratification identified high- and low-risk groups with markedly divergent recurrence-free survival. In the external test set, the prognostic model achieved a C-index of 0.766 and time-dependent AUCs of 0.825 at 2 years and 0.878 at 5 years, showing higher discrimination than BCLC, CNLC, and AJCC staging (all p < 0.001). Performance remained robust across HBV status, histologic differentiation, and BCLC stage.
A radiomics biomarker integrating global tumour phenotype, intratumoral heterogeneity, and preoperative clinical variables enables noninvasive preoperative MVI assessment and may complement morphology-based staging for recurrence risk stratification in HCC.
Authors
Liu Liu, Xie Xie, Yue Yue, Jia Jia, Wang Wang, Liao Liao, Li Li, Yao Yao, Zhang Zhang, Wang Wang, Zhou Zhou, Shi Shi, Ma Ma, Chen Chen, Ma Ma, Zhou Zhou, Yang Yang, Ning Ning, Zhou Zhou, Hu Hu, Li Li, Zhuo Zhuo
View on Pubmed