Transformed common spatial pattern for motor imagery-based brain-computer interfaces.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Motor imagery (MI)-based brain-computer interfaces (BCIs) are popular.
• Common Spatial Pattern (CSP) is effective for decoding MI-related electroencephalogram (EEG) but relies heavily on frequency band selection.
• Previous methods often used filter banks before CSP to address frequency band dependency.

Purpose of the Study:

• To introduce a novel method, transformed Common Spatial Pattern (tCSP), for extracting discriminant EEG features.
• To apply tCSP for feature extraction from multiple frequency bands after CSP.
• To evaluate the effectiveness of tCSP compared to traditional CSP and filter bank CSP (FBCSP).

Main Methods:

• Proposed transformed Common Spatial Pattern (tCSP) algorithm.
• Extracted EEG features from multiple frequency bands post-CSP.
• Tested tCSP on a self-collected dataset and a public BCI competition dataset.
• Conducted an online evaluation of the tCSP method.

Main Results:

• tCSP achieved higher classification accuracy than CSP (approx. 8%) and FBCSP (approx. 4.5%) on the self-collected dataset.
• The combination of tCSP and CSP yielded an average accuracy of 84.77% and a peak of 100%.
• On the BCI competition III dataset, the combined method achieved 94.55% accuracy, outperforming other CSP-based methods.
• Online evaluation showed average accuracies of 80.00% for tCSP and 84.00% for the combined method.

Conclusions:

• Frequency band selection after CSP is superior to selection before CSP for MI-based BCIs.
• tCSP offers a promising approach for decoding MI EEG patterns.
• The findings are significant for advancing BCI development.