Integrating pretreatment CT radiomics and circulating tumor cells using machine learning to predict survival in hepatocellular carcinoma.

Background: Immunotherapy has shown promising potential in the treatment of advanced hepatocellular carcinoma (HCC). This study aimed to establish and validate a multimodal prognostic model based on clinical variables, CT radiomics, and circulating tumor cell (CTC) counts for survival prediction in advanced HCC. Methods: Pretreatment CT images and baseline clinical data were collected from patients with advanced HCC. Radiomic features were extracted from CT images, and candidate machine learning pipelines were screened to derive an optimal radiomics signature. Clinical and biomarker variables associated with overall survival were identified using Cox regression analyses and incorporated into prognostic nomograms. Model discrimination, calibration, and clinical utility were evaluated in internal and external validation cohorts. Results: The addition of immunotherapy was associated with improved prognosis in patients with advanced HCC. Among the prognostic models, the Clinical-Radiomic-CTC nomogram outperformed the Clinical-Radiomic nomogram, showing a higher concordance index (0.789) and higher area under the receiver operating characteristic curves (AUCs) for 1-, 2-, and 3-year OS (0.889, 0.771, and 0.838, respectively). Conclusion: Machine learning models integrating radiomics and CTCs provided robust individualized prognostic prediction, supporting risk stratification and clinical decision-making.