3D Convolution neural network with multiscale spatial and temporal cues for motor imagery EEG classification.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Deep learning methods show promise in motor imagery electroencephalogram (MI EEG) classification.
• Existing methods struggle with low signal-to-noise ratios, inter-subject variability, and neglecting spatial-temporal information.

Purpose of the Study:

• To propose an end-to-end 3D Convolutional Neural Network (CNN) for enhanced 4-class MI EEG classification.
• To address limitations in subject-adaptive feature extraction and capture spatial-temporal dependencies.

Main Methods:

• Developed an end-to-end 3D CNN model for MI EEG signal processing.
• Implemented adaptive weighting for motor-related spatial channels and temporal sampling.
• Extracted multiscale spatial and temporal dependent features.

Main Results:

• Achieved average classification accuracies of 93.06% and 97.05% on two datasets.
• Demonstrated superior performance and robustness across different subjects compared to state-of-the-art methods.
• Successfully applied the method for real-time robot control using MI EEG signals.

Conclusions:

• The proposed 3D CNN effectively extracts multiscale spatial-temporal features for MI EEG classification.
• The method enhances classification accuracy and Brain-Computer Interface (BCI) system performance.
• The approach shows significant potential for practical BCI applications.