Abstract
PURPOSE: To evaluate whether radiomic features from contrast-enhanced computed tomography (CECT) of peritumoral regions can be used to preoperatively predict proliferative hepatocellular carcinoma (PHCC). PATIENTS AND METHODS: Preoperative CT scans from 486 patients with hepatocellular carcinoma (HCC) were retrospectively analyzed and split into training (n = 252), testing (n = 109), and validation (n = 125) cohorts. Radiomic features were extracted from intra- and peritumoral regions (peri-3 mm, peri-5 mm, and peri-10 mm) on arterial phase (AP) and portal venous phase (PVP) images using PyRadiomics. Features were selected with LASSO regression and 10-fold cross-validation, and a radiomics score (Radscore) was calculated as a weighted sum of selected features. Patients were classified into high- and low-risk groups using the optimal Youden's index cutoff. Recurrence-free survival (RFS) was analyzed with Kaplan-Meier curves, feature contributions were quantified using SHapley Additive exPlanations (SHAP), and model performance was assessed by area under the curve (AUC). RESULTS: The Naive Bayes model using peri-5 mm features achieved the highest mean AUC (0.739) and accuracy (0.802), with AUCs of 0.839 and 0.639 in internal and external validation. In the test set, combining intra- and peritumoral features improved the AUC to 0.849 (95% CI: 0.773-0.924; sensitivity: 0.974; specificity: 0.606). In the validation set, AP, PVP, and their combined models achieved AUCs of 0.699, 0.672, and 0.695, respectively. SHAP highlighted in the Naive Bayes model that the increased inhomogeneity of the texture grayscale of the peritumoral tissue in the PVP may be associated with more aggressive HCC subtypes. Multivariable analysis identified rim-APHE (OR = 22.667), mosaic architecture (OR = 5.904), and intratumoral hemorrhage (OR = 4.897) as independent risk factors for PHCC (all p < 0.05). PHCC showed significantly worse RFS than non-PHCC (p < 0.0001). Radscore effectively stratified early recurrence risk (p < 0.0001). CONCLUSION: Radiomic analysis of intratumoral and peri-5 mm enhancement features enables accurate preoperative PHCC identification and may inform intensified postoperative surveillance and adjuvant therapy.