Dual-Phase Computed Tomography-Based Deep Learning Architecture for Three-Year Survival Prediction in Hepatocellular Carcinoma.

Hepatocellular carcinoma (HCC) is a major global health burden, ranking as the sixth most common cancer and the third leading cause of cancer-related mortality. Computed tomography (CT) is widely used for HCC evaluation because of the high spatial resolution and the availability of multiphase acquisition. However, interpretation requires expertise and may be subject to inter-observer variability. The Liver Imaging Reporting and Data System (LI-RADS) standardizes imaging terminology and protocols, recommending arterial and venous phase images for HCC diagnosis. Recent progress in deep learning has significantly advanced computer vision tasks and enabled the development of computer-aided diagnosis (CADx) systems that can improve efficiency and accuracy. In this study, we propose a deep learning framework based on dual-phase CT for predicting three-year survival in patients with HCC. The model is built on MedNeXt with a dual-branch design to capture phase-specific features. A Dual Phase Contextual Fusion Block (DPCFB) enhances cross-phase feature integration, while a Cascaded Damper Block (CDB) incorporates clinical variables and tumor size to improve prognostic modeling. The proposed system achieved predictive performance with an accuracy of 85%, a sensitivity of 83%, a specificity of 86%, and an Area Under the Receiver Operating Characteristic (ROC) Curve (AUC) of 0.88. These findings demonstrate that combining multiphase CT with clinical data supports the accurate prediction of long-term survival in HCC. The proposed framework shows promise as a clinical decision-support tool for prognosis, treatment planning, and patient management.