Hybrid CNN-GRU Models for Improved EEG Motor Imagery Classification.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCIs) leverage electroencephalography (EEG) to interpret neural activity for device control.
• Motor imagery (MI) is a key paradigm for decoding imagined movements within BCI systems.
• Efficient feature extraction from EEG signals is crucial for enhancing classification accuracy and reducing preprocessing demands.

Purpose of the Study:

• To introduce novel hybrid architectures for improved MI classification in EEG-based BCIs.
• To enhance MI classification using data augmentation and a reduced number of EEG channels.
• To evaluate the performance of proposed models against state-of-the-art methods on a public dataset.

Main Methods:

• Development of two hybrid deep learning models: a Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU) and a Convolutional Neural Network-Bidirectional Gated Recurrent Unit (CNN-Bi-GRU).
• Application of data augmentation techniques to improve model robustness.
• Evaluation of models using the publicly available PhysioNet dataset for motor imagery tasks.

Main Results:

• The CNN-GRU classifier achieved peak mean accuracy rates exceeding 99.7% across various motor imagery tasks (left fist, right fist, both fists, both feet).
• The proposed hybrid architectures demonstrated superior performance compared to current state-of-the-art methods.
• Experimental results highlight the efficiency of the models, particularly on small-scale EEG datasets.

Conclusions:

• The developed CNN-GRU and CNN-Bi-GRU architectures provide a faster and more cost-effective solution for user-adaptable MI-BCI applications.
• These models exhibit superior predictive reliability, paving the way for more accessible BCI technology.
• The study underscores the potential of hybrid deep learning approaches for advancing EEG-based BCI performance.