MRI-Based Deep Learning and Radiomics Nomogram for Predicting Hepatocellular Carcinoma Recurrence Within Six Months After Thermal Ablation.

Purpose: Develop a magnetic resonance imaging (MRI)-based deep learning (DL)-radiomics (Rad)-clinical nomogram for predicting early recurrence of hepatocellular carcinoma (HCC) within six months after thermal ablation. Materials and Methods: Barcelona Clinic Liver Cancer (BCLC) stage 0-A HCC patients who underwent dynamic contrast-enhanced MRI before ablation were retrospectively included. Patients were categorized into non early recurrence and early recurrence groups. A clinical model was constructed through logistic regression analysis of clinical information and radiological features. DL score model and Rad score model were developed using DL features and manual features extracted from dynamic contrast-enhanced MRI, with principal component analysis and least absolute shrinkage and selection operator regression methods. The DL-Rad-Clinical nomogram was constructed through logistic regression analysis. The model performance was primarily evaluated using the area under the receiver operating characteristic curve (AUC). Results: A total of 224 patients were included in this study (training set: n = 156; test set: n = 68). The DL-Rad-Clinical nomogram was constructed, including Rad score, DL score, natural logarithm alpha-fetoprotein (LnAFP), and multiple low signal lesions as predictive factors. In the training set, the DL-Rad-Clinical nomogram demonstrated better predictive performance (AUC = 0.896, P < 0.05). In the test set, the DL-Rad-Clinical nomogram had a higher AUC value compared to other models, although the difference was not statistically significant (AUC = 0.774, P > 0.05). Conclusion: The DL-Rad-Clinical nomogram helped in identifying HCC patients with early recurrence within six months following thermal ablation.