Brain Ischemic Stroke Detection through Deep Learning: A Systematic Review on CT vs MRI vs CTA Images

Purpose: Ischemic brain strokes have a high morbidity and death rate, thus it’s vital to obtain a quick diagnosis and imaging. Computer-aided diagnosis (CAD) has become popular in medical imaging and diagnostic radiology research. In the field of medical image analysis, deep learning (DL) approaches have recently shown greater performance over earlier, more advanced machine learning techniques. Acute Ischemic stroke (AIS) is one of the medical sectors where DL has conducted substantial research. The systematic review examines the performance of deep learning models across different imaging modalities, highlighting their strengths and limitations in identifying ischemic strokes. Key aspects such as sensitivity, specificity, and overall accuracy are assessed, providing insights into the comparative effectiveness of CT, MRI, and CTA in stroke detection. In contrast with other reviews in this domain, this paper offers a concise summary of the most notable DL methods applied in the classification, detection, and segmentation of acute ischemic brain stroke, focusing on popular imaging techniques like Computed Tomography (CT), Magnetic Resonance Imaging (MRI), and CT Angiography (CTA). This survey also highlights datasets and data acquisition challenges and attempts to provide a comprehensive overview of data preprocessing, as well as insight into publicly available datasets. Methods and Results: This study aims to give an idea of how training and testing datasets should be handled by evaluat- ing recent studies. This review discusses the challenges associated with each imaging modality, including image noise, artifacts, and variability in acquisition protocols. Strategies to mitigate these challenges through preprocessing techniques and model optimization are explored, aiming to improve the robustness and reliability of deep learning-based stroke de- tection systems. Moreover, this research contains a brief discussion of recent deep learning architectures and an analysis of performances. Conclusions: Overall, this systematic review contributes to the understanding of current advancements in brain ischemic stroke detection through deep learning, offering valuable insights for researchers and clinicians seeking to leverage these technologies for improved patient outcomes. Future directions and potential research avenues are also discussed to guide further advancements in this critical area of medical imaging and diagnosis.