Peri-ablational contrast-enhanced CT radiomics for predicting early progression of hepatocellular carcinoma after microwave ablation: development and external validation of a clinical-radiomics model.
Authors
Affiliations (5)
Affiliations (5)
- Department of Ultrasound,Central Hospital,Tianjin University/Tianjin Third Central Hospital, Tianjin, China.
- Tianjin Key Laboratory of Extracorporeal Life Support for Critical Diseases, Tianjin Artificial Cell Engineering Technology Research Center, Tianjin Institute of Hepatobiliary Disease, Tianjin, China.
- Department of Ultrasound, Tianjin Second People's Hospital, Tianjin, China.
- Department of Ultrasound,Central Hospital,Tianjin University/Tianjin Third Central Hospital, Tianjin, China. [email protected].
- Tianjin Key Laboratory of Extracorporeal Life Support for Critical Diseases, Tianjin Artificial Cell Engineering Technology Research Center, Tianjin Institute of Hepatobiliary Disease, Tianjin, China. [email protected].
Abstract
To develop and externally validate a peri-ablational contrast-enhanced CT radiomics model for predicting early hepatocellular carcinoma (HCC) progression after microwave ablation (MWA) and to assess its incremental value over clinical predictors. This retrospective multicenter study included 376 patients in training (n = 235), internal validation (n = 60), and external validation (n = 81) cohorts. Early progression was defined as local tumor progression or new intrahepatic HCC within 1 year after MWA. Radiomics features were extracted from 5-mm peri-ablational regions on multiphasic CT obtained 1-2 months after MWA. Six machine-learning algorithms and phase combinations were evaluated. The selected radiomics model was combined with independent clinical predictors. Discrimination, calibration, clinical utility, and model interpretability were assessed. The arterial-plus-portal venous phase (AP + PP) multilayer perceptron model achieved AUCs of 0.857, 0.815, and 0.903 in the training, internal validation, and external validation cohorts, respectively, with performance comparable to the three-phase model. Log<sub>10</sub> AFP and tumor size > 2 cm were independent clinical predictors. The combined clinical-radiomics model achieved AUCs of 0.926, 0.884, and 0.963, respectively. In external validation, sensitivity, specificity, and accuracy were 84.2%, 96.8%, and 93.8%, respectively. SHAP analysis identified the radiomics score as the most influential predictor. Peri-ablational AP + PP CT radiomics combined with AFP and tumor size demonstrated good discrimination for early HCC progression after MWA and may facilitate individualized risk assessment, pending further validation.