Application of nninteractive Automatic Segmentation in Habitat Analysis of Contrast-Enhanced CT for Predicting Microvascular Invasion of Hepatocellular Carcinoma

Objective To quantify hepatocellular carcinoma (HCC) intratumoral heterogeneity using preoperative contrast‑enhanced CT‑based habitat imaging and to develop a nomogram for preoperative prediction of microvascular invasion (MVI) risk. Methods In this retrospective study, HCC patients who underwent resection (January 2020–May 2024) were included. A deep‑learning model segmented 3D tumor volumes on arterial‑phase CT. k‑means clustering (k = 3) partitioned tumors into distinct habitat subregions, from which habitat‑specific and whole‑tumor radiomic features were extracted. Patients were randomly divided 7:3 into training and internal validation cohorts. A multivariable logistic regression model integrating significant imaging and clinicopathological variables was constructed to develop an MVI‑predictive nomogram. Results Among 206 HCC patients (78 MVI‑positive, 128 MVI‑negative), the habitat model showed excellent predictive performance (AUC = 0.812, 95% CI: 0.698–0.927), outperforming clinical (AUC = 0.671) and radiomics models (AUC = 0.765) in the validation set, though pairwise comparisons were not statistically significant (all P > 0.05). The combined model—integrating habitat features, radiomics score, and clinical predictors—achieved the highest AUC (0.847, 95% CI: 0.749–0.945) and was significantly better than the clinical model (P = 0.006). Hosmer‑Lemeshow tests confirmed good calibration in both training (P = 0.268) and validation (P = 0.644) sets. Conclusion A CT‑based habitat imaging nomogram supported by automated segmentation enables accurate, noninvasive preoperative prediction of MVI risk, offering a practical tool to guide individualized surgical and postoperative management in HCC.