Background: To establish and vertify a nomogram model that integrates multiparametric magnetic resonance imaging (MRI) radiomic signatures and clinical features to predict satellite nodules (SNs) and recurrence-free survival (RFS) in hepatocellular carcinoma (HCC) patients.
Methods: Data from 244 patients with HCC who underwent multiparametric MRI were analyzed and randomly assigned into a training (n = 170) dataset and a validation dataset (n = 74). A support vector machine algorithm was employed to develop T1WI (T1-weighted imaging), T2WI (T2-weighted imaging), arterial phase (AP), portal-venous phase (PVP), and integrated MRI radiomic models. The selected signatures were combined with independent clinical factors to construct a nomogram model. The performance of these predictive models in the prediction of SNs and RFS was assessed with the AUC and Kaplan-Meier analysis, respectively.
Results: Portal vein tumor thrombosis and peritumoral enhancement were significant clinical indicators of SNs (P < 0.05). In predicting SNs, the nomogram model demonstrated the highest AUC value of 0.94 in the training dataset and 0.83 in the validation dataset. This was followed by the integrated MRI (0.93 and 0.79), AP (0.92 and 0.82), T2WI (0.91 and 0.78), PVP (0.90 and 0.80), and T1WI models (0.88 and 0.77). Compared with SNs (-) patients, SNs (+) patients had a significantly lower median RFS (61.3 vs. 18.6 months, P < 0.001). Additionally, nomogram predicted SNs (+) had a lower median RFS compared to SNs (-) (20.5 vs. 63.1 months, P < 0.001).
Conclusion: The nomogram model based on multiparametric MRI radiomics signatures demonstrated substantial efficacy in predicting SNs and RFS in patients with HCC.
Keywords: Hepatocellular carcinoma; Magnetic resonance imaging; Radiomics; Recurrence-free survival; Satellite nodules.
Copyright © 2025 Elsevier Inc. All rights reserved.