Abstract: We developed a Clinical-Radiomics model (CRM) to predict early hematoma expansion in supratentorial spontaneous intracerebral hemorrhage (ssICH). A retrospective cohort of 307 ssICH patients was split into training (n=214) and validation (n=93) sets. This model integrated clinical factors (GCS score, serum ionized calcium, onset time to non-contrast CT scan) and imaging markers (blend sign, midline shift, radiomics score). The boruta algorithm selected 15 core radiomics features. Among seven machine learning models, random forest achieved optimal performance (training AUC=0.933, validation AUC=0.936). Decision curve analysis demonstrated a higher net benefit strategy at 0-20% threshold, with 91% sensitivity for OT≤2h patients. Our CRM enables precise hematoma heterogeneity quantification and supports emergency tiered interventions.