Motor imagery recognition with automatic EEG channel selection and deep learning.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Modern brain-computer interfaces (BCIs) rely on numerous electroencephalogram (EEG) channels for motor imagery (MI) detection.
• Selecting relevant EEG channels and extracting discriminative features is crucial for enhancing BCI control capabilities.
• Irrelevant or redundant EEG channels can degrade classification accuracy and system robustness.

Purpose of the Study:

• To propose and validate a deep learning-based approach for automatic EEG channel selection (ACS) in MI tasks.
• To develop a method for recognizing two distinct MI states by identifying and utilizing the most relevant EEG channels.
• To improve the accuracy and robustness of MI classification compared to traditional methods.

Main Methods:

• Utilized a sparse squeeze-and-excitation module to derive channel weights based on their contribution to MI classification.
• Developed an automatic channel selection (ACS) strategy integrated with a convolutional neural network (CNN).
• The CNN was designed to effectively extract and leverage time-frequency features from selected EEG channels.

Main Results:

• Achieved an average classification accuracy of 87.2±5.0% in discriminating between right-hand and feet MI movements.
• Demonstrated a 37.3% performance improvement over a state-of-the-art channel selection approach.
• Experiments were conducted with 25 healthy subjects performing MI tasks.

Conclusions:

• The proposed ACS method offers significant advantages over fixed channel configurations, reducing computational complexity.
• Selecting fewer, relevant EEG channels enhances MI classification performance and system efficiency.
• This approach facilitates the development of more natural, real-time BCIs for controlling robotic devices.