Abstract
BACKGROUND AND AIMS: Accurate preoperative prediction of microvascular invasion (MVI) in hepatocellular carcinoma (HCC) is crucial for treatment planning. This study aimed to develop and validate a multi-phase magnetic resonance imaging (MRI)-based radiomics model for predicting MVI in HCC patients. METHODS: This retrospective study included 110 HCC patients (training: n = 77; validation: n = 33) who underwent preoperative multi-phase MRI. Radiomics features were extracted from four MRI phases (non-contrast, arterial, portal, and hepatobiliary). Feature selection was performed using least absolute shrinkage and selection operator regression, and five machine learning classifiers were evaluated. Model performance was assessed using standard metrics including area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy. RESULTS: The four-phase radiomics model with logistic regression classifier showed optimal performance in both the training (AUC = 0.896; 95% confidence interval, 0.792-0.963) and validation cohorts (AUC = 0.889, 95% confidence interval, 0.781-0.982), outperforming the single-phase (AUC = 0.789), two-phase (AUC = 0.815), and three-phase models (AUC = 0.848) in the validation cohort. In the validation cohort, the model achieved balanced performance with sensitivity, specificity, accuracy, and precision all reaching 0.857. CONCLUSIONS: The multi-phase MRI-based radiomics model significantly improves MVI prediction accuracy in HCC patients. This non-invasive approach could enhance preoperative assessment and treatment planning.