Deep Learning based early detection of Hematoma Expansion from Non-Contrast CT scan of Brain in Intracerebral Hemorrhagic Stroke

Hematoma expansion in spontaneous intracerebral hemorrhage (ICH) significantly impacts patient outcomes, often requiring urgent hospitalization and surgical intervention. This study presents a deep learning-based approach for detecting hematoma enlargement in non-contrast CT (NCCT) scans using segmentation masks. A ResNet-18 classifier was trained on hematoma masks to predict expansion likelihood, while the ABC/2 method quantified volume changes and percentage growth. The model achieved 80% accuracy, using intervention thresholds of <10 mL for medical therapy and >30 mL for surgery. Future work will enhance performance with multimodal data fusion, 3D CNNs, and explainable AI tools like Grad-CAM, supporting real-time clinical application.