Habitat and peritumoral CT radiomics accurately predict early treatment response to hepatic arterial infusion chemotherapy combined with tyrosine kinase inhibitors and programmed death-1 inhibitors in unresectable hepatocellular carcinoma.

Background: The pronounced heterogeneity of hepatocellular carcinoma (HCC) leads to variable responses to hepatic arterial infusion chemotherapy combined with tyrosine kinase inhibitors and PD-1 inhibitors (HTP therapy). This study aimed to develop and validate a combined model integrating habitat radiomics, peritumoral radiomics, and clinical factors to predict early treatment response in patients with unresectable HCC receiving HTP therapy. Methods: This retrospective two-center study enrolled 213 patients with unresectable HCC who received HTP therapy. Patients from Center 1 (n = 143) were randomly allocated to a training cohort (n = 100) and an internal validation cohort (n = 43); patients from Center 2 (n = 70) served as an independent external testing cohort. Regions of interest from portal venous phase images were partitioned into three habitat subregions via K-means clustering and expanded at thicknesses of 3 mm and 9 mm. Following feature extraction and selection, multiple machine learning algorithms were compared. A combined model was constructed by integrating the habitat model, the best peritumoral model, and clinical risk factors. Results: Among the 213 treated patients, 109 achieved complete or partial response, yielding an objective response rate of 51.2%. The Habitat model demonstrated superior predictive performance compared to other single models, with AUCs of 0.921, 0.883, and 0.826 in the training, internal validation, and external testing cohorts. The combined model yielded even higher AUCs of 0.955, 0.968, and 0.893 across the three cohorts. Decision curve analysis and calibration curves further confirmed that the combined model provided high net clinical benefit and exhibited good calibration. Conclusion: The combined model, integrating habitat and peritumoral radiomics with clinical factors, accurately predicts early treatment response to HTP therapy in unresectable HCC. Moreover, the model-derived risk stratification significantly differentiated progression-free survival, underscoring its potential prognostic utility and offering a non-invasive tool to facilitate personalized treatment decisions.