Abstract
OBJECTIVES: To establish and validate a multi-parameter model for the prediction of early recurrence in patients with hepatitis B-associated hepatocellular carcinoma (HBV-HCC) after microwave ablation. METHODS: This study retrospectively reviewed the clinical features and preoperative magnetic resonance imaging (MRI) scans of 166 patients with HBV-HCC who underwent microwave ablation at two hospitals. The training cohort comprised 116 patients from the first hospital (n = 116; mean age, 56 years; 84 male patients), while 50 patients from the second hospital constituted the external validation cohort (n = 50; mean age, 60 years; 38 male patients). A transformer-based deep learning network was used to fuse images from multi-sequence MRI and predict recurrence within 1 year after microwave ablation. Additionally, a nomogram based on deep learning radiomics and clinical features was developed and externally validated in a validation group from a second hospital. RESULTS: The combined model was better than the clinical model and MRI model in predicting early recurrence of hepatitis B-associated hepatocellular carcinoma within 1 year after microwave ablation. Nomograms based on joint models include aspartate aminotransferase, portal hypertension, and deep learning-based radiomics scores. The areas under curves of the models in the training group and the validation group were 0.868 (95% CI: 0.793-0.924) and 0.842 (95% CI: 0.711-0.930), respectively, indicating high prediction ability. The results of decision curve analysis showed that the combined model had good clinical application value and correction effect. CONCLUSIONS: Our nomogram combined with clinical features and preoperative magnetic resonance imaging features effectively predicted early recurrence of hepatitis B-associated hepatocellular carcinoma within 1 year after microwave ablation.