MRI-based habitat radiomics and deep learning for predicting vessels encapsulating tumor clusters and survival in hepatocellular carcinoma.

OBJECTIVE: The study sought to develop and validate an MRI-based deep learning radiomics (DLR) nomogram for preoperative prediction of vessels encapsulating tumor clusters (VETC) and recurrence-free survival (RFS) in hepatocellular carcinoma (HCC). MATERIALS AND METHODS: The dual-center study retrospectively enrolled 625 HCC patients who underwent preoperative Gd-EOB-DTPA-enhanced MRI, including training (n = 296), internal (n = 126), and external (n = 203) test sets. Clinical-radiologic characteristics were selected to develop a clinical-radiologic model. Habitat radiomics and deep learning (DL) features were extracted and selected to develop the habitat radiomics and DL models using the machine learning classifiers. The DLR nomogram model was ultimately constructed by integrating univariate-selected clinical-radiologic characteristics with habitat radiomics and DL scores. Both univariable and multivariable Cox regression analyses were performed to identify independent prognostic factors and develop a prognostic model for RFS. RESULTS: In the external test set, the DLR nomogram model yielded a higher area under the curve (AUC) than the clinical-radiologic model (0.752 vs 0.678; p = 0.004), while habitat radiomics (0.750) and DL models (0.748) showed comparable performance (both p > 0.05). The DLR nomogram consistently demonstrated the higher F1-scores across all three sets. The prognostic model incorporating AFP (hazard ratio (HR), 1.628 [95% CI: 1.113-2.380]; p = 0.012) and DLR score (1.279 [1.051-1.557]; p = 0.014) achieved C-indexes of 0.679 and 0.642 for RFS in the internal and external test sets. CONCLUSION: The DLR nomogram model helps predict VETC in HCC and assess the risk for RFS. CRITICAL RELEVANCE STATEMENT: Interpretable deep learning radiomics nomogram model provides clinicians with more precise technical support for preoperative prediction of VETC status and RFS in HCC, potentially aiding in clinical decision-making and follow-up strategies. KEY POINTS: Vessels encapsulating tumor clusters (VETC) is a critical predictor of aggressive hepatocellular carcinoma. The deep learning radiomics (DLR) nomogram model helps predict VETC, and the DLR score serves as an independent prognostic factor for recurrence-free survival. The model demonstrated favorable interpretability through the SHAP method.