An in-depth survey on Deep Learning-based Motor Imagery Electroencephalogram (EEG) classification

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Electroencephalogram (EEG)-based Brain-Computer Interfaces (BCIs) enable communication between the brain and external devices.
• Motor Imagery (MI) is a common BCI paradigm with applications in medicine and smart homes.
• Classifying MI-EEG signals is challenging due to low signal-to-noise ratio (SNR) and non-stationarity.

Purpose of the Study:

• To systematically survey Deep Learning (DL) based methods for MI-EEG classification.
• To provide guidelines for fair performance comparison of DL models.
• To analyze the impact of network architecture on classification performance.

Main Methods:

• Comprehensive review of DL-based MI-EEG classification techniques, including input formulations and network architectures.
• Fair evaluation of representative DL models using author-released source code.
• Ablation studies on network architectures to understand component contributions.

Main Results:

• Effective feature fusion is crucial for multi-stream CNN models.
• Combining LSTM with spatial feature extraction enhances classification.
• Dropout has minimal impact, and fully connected layers may not improve performance despite increased parameters.

Conclusions:

• Deep learning advances significantly aid MI-EEG classification.
• Guidelines are provided for future research to ensure fair model performance comparison.
• Open issues and future research directions in MI-EEG classification are highlighted.