ARCHIVES

Brain strokes, medically referred to as cerebrovascular conditions, are identified as one of the top causes of mortality among diverse age groups globally. Acute ischemic strokes and hemorrhagic stroke attacks nearly 12-13 million people on average globally per year. Early detection and classification of stroke types is pivotal for timely medical intervention. The traditional methods primarily rely on Magnetic Resonance Imaging (MRI) and Diffusion-Weighted Imaging (DWI). This method imposes severe drawbacks, including false results that can be inaccurate, time sensitivity and other anomalies that could lead to significant problems if overlooked. With advancements in technology and increased efficiency of machine learning models, deep learning methodologies have significantly revolutionised stroke detection through extensive exploitation of multiple neural networks. Considerable optimisation techniques have been developed to facilitate rapid image analysis, provide accurate and precise findings, segment lesions, aid in diagnosis, and support treatment. This comprehensive paper discusses efficient deep learning methodologies and paradigms, such as Salp Shuffled Shepherd optimisation (S3O) [1] for deriving suitable features from pre-processed images, Vision Transformers(ViT)[2] for attention-based diagnosis, EfficientNet[7] for scalable extraction and classification, and a few other techniques for automated brain stroke detection. The accuracy rates of each model are predicted and validated through analysis on different publicly available datasets, hence making the models pre-trained. Each model demonstrates a highly impressive reliability of over 95%, while still leaving room for advancement. The paper also identifies implementation barriers and proposes evidence-oriented solutions to improve diagnostic equity and improve recovery prospects.