Baseline PET/CT Radiomics and Clinical Indicators for Pretreatment Response Prediction in Unresectable Hepatocellular Carcinoma: Model Development and Internal Validation.

Abstract

IntroductionPretreatment prediction of benefit from targeted therapy plus immunotherapy remains uncertain in unresectable hepatocellular carcinoma (HCC). We developed and internally validated a model integrating baseline 18F-FDG PET/CT radiomics with clinical indicators.MethodsThis single-center retrospective study included 325 patients, comprising a development cohort (n = 182) and an internal validation cohort (n = 143). Clinical/PET, radiomics, and fusion logistic models were developed to predict 12-week objective response assessed according to the modified Response Evaluation Criteria in Solid Tumors. Development-derived preprocessing parameters, coefficients, and cutoffs were applied unchanged to validation. Discrimination was assessed using AUCs with bootstrap 95% confidence intervals and paired patient-level bootstrap comparisons; calibration was assessed using the Brier score, calibration intercept, and calibration slope.ResultsIn the 143-patient internal validation cohort, the clinical, radiomics, and fusion models had AUCs of 0.69 (0.60-0.77), 0.75 (0.67-0.83), and 0.84 (0.78-0.90), respectively. The paired AUC difference was 0.16 (95% CI, 0.05 to 0.26; P = 0.002) versus the clinical model and 0.09 (-0.001 to 0.19; P = 0.053) versus the radiomics model. The fusion-model Brier score was 0.16, calibration intercept was -0.02, and calibration slope was 0.83.ConclusionThe fusion model integrating clinical and radiomics features demonstrated favorable discrimination and calibration for predicting 12-week objective response in patients with HCC. It improved predictive performance compared with the clinical model alone and may provide a noninvasive tool for pretreatment response assessment. Prospective multicenter studies are warranted to evaluate its generalizability and clinical utility.