FBMSNet: A Filter-Bank Multi-Scale Convolutional Neural Network for EEG-Based Motor Imagery Decoding.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Motor imagery (MI) is crucial for brain-computer interfaces (BCIs).
• Decoding MI intentions from electroencephalography (EEG) is challenging due to complex brain patterns and limited data.
• Existing methods struggle with the inherent complexity and small sample sizes in MI decoding.

Purpose of the Study:

• To propose an end-to-end deep learning model, Filter-Bank Multiscale Convolutional Neural Network (FBMSNet), for improved MI classification.
• To enhance the accuracy and robustness of EEG decoding for BCI applications.

Main Methods:

• Utilized a filter bank for multiview spectral representation of EEG data.
• Applied mixed depthwise convolution for multi-scale temporal feature extraction.
• Incorporated spatial filtering to reduce volume conduction effects.
• Employed joint supervision of cross-entropy and center loss for feature optimization.

Main Results:

• FBMSNet achieved 79.17% accuracy for four-class MI classification and 70.05% for two-class classification on benchmark datasets.
• Significantly outperformed existing state-of-the-art EEG decoding methods.
• Demonstrated superior performance on the BCI Competition IV 2a and OpenBMI datasets.

Conclusions:

• FBMSNet effectively improves EEG decoding performance for MI tasks.
• The proposed model shows significant potential for developing more robust BCI applications.
• Source code is publicly available for further research and development.