An improved model using convolutional sliding window-attention network for motor imagery EEG classification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor imagery-based electroencephalogram (MI-EEG) is crucial for brain-computer interfaces (BCIs) and neural rehabilitation.
• Existing MI-EEG models face challenges in spatiotemporal feature extraction, learning, and dynamic selection.

Purpose of the Study:

• To address limitations in current MI-EEG classification models.
• To introduce a novel deep learning architecture for enhanced MI-EEG signal processing.

Main Methods:

• Proposed a convolutional sliding window-attention network (CSANet).
• Incorporated novel spatiotemporal convolution, sliding window mechanisms, and two-stage attention blocks.
• Utilized multi-scale spatiotemporal feature extraction and adaptive selection.

Main Results:

• CSANet achieved superior performance over state-of-the-art models on BCI-2a and Physionet MI-EEG datasets.
• Demonstrated significant improvements in classification accuracy: 4.22% within-individual and 2.02% between-individual.
• Validated the effectiveness of attention mechanisms and sliding windows in feature learning and selection.

Conclusions:

• CSANet offers a novel and accurate classification model for MI-EEG BCIs.
• The model provides a feasible scheme for neural rehabilitation assessment, particularly for Parkinson's and stroke.
• The findings support the advancement and application of MI-EEG in clinical settings.