Hemorrhagic stroke, a leading cause of mortality and disability worldwide, traditionally relies on manual delineation of key brain regions and hematomas in CT images by physicians, which is time-consuming and labor-intensive. Recent advances in deep learning-based computer-aided diagnostic techniques have improved segmentation efficiency. However, the similar pixel intensity between the basal ganglia and thalamus in CT images, as well as the blurred edges of hematomas, pose significant challenges for precise segmentation. To address these issues, we propose a Mask-Focused Edge-Perceptive Dual Distillation Method, which distills the CT brain region and hematoma feature knowledge from a medical vision large model fine-tuned on our private dataset into a lightweight student model. This method aligns teacher-student features, focuses on target regions using masks to accurately distinguish the basal ganglia from the thalamus, and further enhances segmentation precision through edge-perceptive loss. Experimental results demonstrate that our method achieves rapid and accurate performance in brain region and hematoma segmentation tasks.