Abstract Stroke is a leading cause of death and long-term disability worldwide. Early and accurate diagnosis is crucial for effective treatment and improved patient outcomes. Electroencephalography (EEG) has emerged as a promising non-invasive method for stroke detection, valued for its cost-effectiveness, portability, and real-time monitoring of brain activity. However, EEG-based diagnosis faces considerable challenges, including high-dimensional and non-stationary signals, limited data availability, and the computational complexity of deep learning models. This paper reviews advanced deep learning techniques for EEG-based stroke detection and classification. It reviews various architectures, including convolutional neural networks (CNNs), recurrent neural networks (RNNs), and hybrid models such as CNN-LSTM. The paper also discusses deep metric learning approaches, like Weighted Convolutional Siamese Networks (WCSN), focusing on their effectiveness and clinical applicability. The study highlights key challenges, such as limited high-quality labeled EEG datasets, difficulties in model interpretability, and the high computational cost of deep learning diagnostics. Furthermore, it explores potential advancements to enhance clinical feasibility. These include multimodal imaging integration (EEG combined with MRI and CT), computational optimization, and explainable AI. By consolidating recent findings, this paper provides a roadmap for future research, aiming to bridge the gap between theoretical models and clinically viable tools for early, effective stroke diagnosis. Graphical Abstract
Ghiyabi et al. (Thu,) studied this question.