CT-based habitat imaging integrated with radiomics and clinicopathology for noninvasive prediction of microvascular invasion in hepatocellular carcinoma

Objective To develop and validate a CT-based habitat imaging model incorporating intratumoral microenvironment heterogeneity analysis for noninvasive preoperative prediction of microvascular invasion (MVI) in hepatocellular carcinoma (HCC), addressing limitations of conventional radiomics in characterizing intra-tumor heterogeneity. Methods This retrospective study included 216 patients with pathologically confirmed HCC undergoing resection. Preoperative portal venous phase contrast-enhanced CT and corresponding postoperative histopathologic MVI status were collected. Habitat (functional heterogeneous subregion) and conventional radiomic features were extracted. A habitat risk score and radiomic score were calculated. Clinical-pathologic factors (Edmondson grade, p53/CD10 expression, tumor diameter) identified via univariate/multivariate analyzes were integrated into a predictive model visualized as a nomogram. Performance was assessed by AUC, calibration curves, and decision curve analysis (DCA). DeLong’s test compared ROC curves. Results In the training set, the combined model achieved an AUC of 0.862 (95% CI: 0.797–0.926); in the validation set, AUC was 0.814 (95% CI: 0.710–0.918), both significantly outperforming individual models (all P < 0.05). Calibration curves showed good agreement (Hosmer–Lemeshow P = 0.60). DCA indicated net benefit at thresholds of 15%–65%. DeLong’s test confirmed the combined model had higher AUC than the clinical model (Z = –3.21, P < .05) and radiomics-only model (Z = –2.05, P < .05). Conclusion The CT-based habitat imaging model quantified intratumoral heterogeneity and, combined with clinicopathologic data, provided a reliable noninvasive tool for preoperative MVI risk stratification in HCC, with strong clinical potential.