Enhancing hematoma expansion prediction in hypertensive intracerebral hemorrhage based on habitat and perihematomal edema radiomics from non-contrast CT: a dual-center study.

Objectives: Characterizing the microenvironmental habitats within the hematoma may yield crucial imaging biomarkers and improve the early prediction of hematoma expansion (HE) in patients with hypertensive intracerebral hemorrhage (HICH). Our objective was to construct and validate a combined model that integrates clinical data with whole-hematoma radiomics, habitat radiomics of the hematoma, and perihematomal edema (PHE) radiomics features extracted from non-contrast computed tomography (NCCT) images for preoperative HE prediction. Methods: This retrospective dual-center cohort of 353 HICH patients. Based on baseline NCCT images, radiomics features were extracted from the whole hematoma, three distinct habitats within the hematoma, and the PHE region. Five models were constructed: a clinical model, a whole-hematoma radiomics model, a habitat-based radiomics model, a PHE radiomics model, and a combined model. Model performance was evaluated using receiver operating characteristic (ROC) curve analysis. Results: The combined model integrated with smoking history, island sign, maximum distance of the PHE, and the whole-hematoma, habitat, and PHE radiomics models, achieved the best predictive performance. In the training, testing, and validation sets, the combined model predicted the area under the curve for HE as 0.951 (95% CI: 0.915-0.986), 0.937 (95% CI: 0.883-0.991), and 0.939 (95% CI: 0.888-0.989), respectively. Conclusion: The NCCT-based combined model integrating clinical data, whole-hematoma radiomics, habitat radiomics, and PHE radiomics improves HE prediction in patients with HICH, providing a noninvasive tool with potential for guiding treatment strategies.