Stroke remains a major cause of morbidity and mortality worldwide, with India ranking fourth in stroke-related deaths, reporting 53.5 deaths per 100,000 people. Despite significant advances in high-income countries, low- and middle-income regions such as India continue to face diagnostic delays and limited imaging access. Artificial intelligence (AI) has shown potential to improve diagnostic speed and accuracy, particularly in low-resource settings where specialist availability and timely interpretation remain barriers to acute stroke management.

A total of 705 records were identified (PubMed = 23; Scopus = 682). Following screening, 48 studies met inclusion criteria. Most used deep learning models for CT-based detection or segmentation. This includes convolutional neural networks (CNNs), U-Net variants, and hybrid machine learning ensembles. Studies reported high diagnostic accuracy across both ischemic and hemorrhagic stroke detection tasks, with models achieving sensitivities ranging from 90–98% on institutional and public datasets. However, generalizability remains limited due to small sample sizes, heterogeneity in data sources, and lack of multicenter clinical validation.