Abstract
Detecting intracranial hemorrhage promptly and accurately is vital for ensuring optimal patient outcomes, especially in areas with limited access to radiologists. This paper introduces an automated method employing advanced convolutional neural networks (CNNs) to identify intracranial hemorrhage from Computed Tomography (CT) scans. The methodology involves comprehensive preprocessing techniques tailored for CT scan analysis. Subsequently, we systematically evaluate various CNN models, including DenseNet, ResNext101, and EfficientNet. Remarkably, InceptionNet emerges as the most proficient model, achieving an accuracy rate of 84.03%. Our optimized DenseNet model demonstrates exceptional performance in accurately detecting intracranial hemorrhage and categorizing its subtype. By leveraging state-of-the-art machine learning techniques, our method facilitates swift and precise identification of hemorrhagic conditions from CT scans, empowering healthcare professionals to make informed decisions and provide timely interventions. This advancement holds significant promise for improving patient care, particularly in underserved regions where radiologist shortages persist.