Efficient Brain Hemorrhage Segmentation: A Compact 3D U-Net Deep Learning Approach

Brain hemorrhage, also known as an intracranial hemorrhage, is a major health concern that necessitates prompt and precise diagnosis for effective management. This study presents an advanced approach to segmenting brain hemorrhage by utilizing 3D brain Computed Tomography (CT) scans and Deep Learning (DL) techniques. The proposed framework leverages 3D U-Net, 3D ResU-Net, and 3D Convolutional Neural Networks (CNNs) to capture spatial and contextual information for precise segmentation. Patch-based training was employed to efficiently handle high-resolution volumetric data and elevate the model's capability to identify hemorrhagic regions. The finding from the experiment indicate that the 3D U-Net model outperforms other approaches, across evaluation standards and achieved the accuracy of 96.6%. These highlight the potential of 3D U-Net for accurate and automated brain hemorrhage detection, offering significant support for clinical decisionmaking and early intervention. This study emphasizes the contribution of DL to the progress of medical imaging for critical neurological disorders.