A Comprehensive Review of Convolutional Neural Networks for Motor Imagery EEG Analysis in Brain–Computer Interface Systems

Recent advancements in biomedical signal processing have significantly advanced Brain–Computer Interface (BCI) technologies, allowing more direct communication between the human brain and external devices. Among the available neuroimaging techniques, Electroencephalography (EEG) has become particularly valuable due to its non-invasive nature, high temporal resolution, and ability to capture brain dynamics during cognitive and motor tasks. This paper provides an in-depth review of Convolutional Neural Network (CNN)-based approaches for analyzing Motor Imagery (MI) EEG signals in BCI systems. It traces the development of preprocessing, feature extraction, and classification techniques, highlighting the ongoing shift from handcrafted feature design toward deep learning–based automatic feature extraction. The reviewed studies are organized into nine categories of feature extraction methods, and the performance of CNN-based, RNN-based, and hybrid architectures is systematically examined. The findings indicate that CNN-RNN hybrid models often achieve superior performance by effectively capturing spatial and temporal dependencies in EEG signals, while CNNs remain the leading choice for spatial-spectral representation learning. Despite notable progress, issues such as intersubject variability, low signal-to-noise ratio, and limited generalization across datasets continue to present challenges. This review outlines current trends, identifies persistent limitations, and suggests potential directions for future work to guide the development of more accurate, adaptive, and real-time EEG-based BCI systems for both clinical and non-clinical applications.

کلیدواژه ها:

Motor Imagery EEG ، Brain–Computer Interface ، Convolutional Neural Networks ، Signal Classification