SEGMENTATION OF SUBARACHNOID HEMORRHAGE ON BRAIN CT IMAGES USING U-NET AND ATTENTION U-NET: A COMPARATIVE ANALYSIS

Subarachnoid Hemorrhage (SAH) represents a critical medical condition resulting from bleeding in the subarachnoid space, typically due to the rupture of an aneurysm or trauma. Timely identification is vital to avoid long-term neurological impairment. This research assesses the efficacy of U-Net compared to Attention U-Net for the segmentation of SAH in brain CT images, aiming to determine if attention mechanisms enhance segmentation precision. The motivation for this comparison stems from the clinical difficulty in identifying subtle or low-contrast hemorrhagic areas that traditional architectures like U-Net might miss; in contrast, attention-based models are constructed to capture spatial details more proficiently. Both architectures were evaluated using a publicly available SAH CT dataset and assessed on metrics including Dice Score, Intersection over Union (IoU), Precision, Recall, and F1 Score. Attention U-Net outperformed U-Net with higher scores of Dice (0.896) and IoU (0.877), whereas U-Net excelled in precision. Visual assessments also indicated that Attention U-Net was superior in delineating diffuse hemorrhagic regions. These findings advocate for the incorporation of attention mechanisms to enhance segmentation accuracy and clinical relevance in neuroimaging