A neural network-based optimal spatial filter design method for motor imagery classification

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Spatial filtering is crucial for feature extraction in motor imagery-based brain-computer interfaces (BCIs).
• Conventional methods like Common Spatial Patterns (CSP) have limitations in optimizing features for accurate classification.

Purpose of the Study:

• Introduce a novel spatial filter network (SFN) for enhanced motor imagery signal classification.
• Address shortcomings of CSP by simultaneously optimizing spatial filters and classifiers.

Main Methods:

• Developed a two-layer neural network (SFN) with a spatial filtering layer and a classifier layer.
• Employed non-linear mapping functions to link layers and optimize filters and classifiers simultaneously.
• Modified feed-forward structures and derived forward/backward equations for motor imagery EEG signal classification.

Main Results:

• SFN effectively maximizes between-class variance while minimizing within-class variance.
• Demonstrated superior performance compared to conventional CSP and one-versus-rest CSP on BCI competition datasets.
• Achieved increased classification accuracy for motor imagery EEG signals.

Conclusions:

• SFN presents a robust alternative for motor imagery EEG signal classification.
• Simultaneous optimization of spatial filters and classifiers leads to improved BCI performance.
• The proposed method offers enhanced feature extraction and classification accuracy in BCIs.