Improving pre-movement pattern detection with filter bank selection.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Pre-movement decoding of electroencephalography (EEG) signals is crucial for detecting action onsets before upper limb movement.
• Low-frequency EEG signals are commonly used for this purpose.
• Accurate decoding enables non-invasive control of external devices for individuals with motor disabilities.

Purpose of the Study:

• To propose and evaluate a novel binary classification method, filter bank standard task-related component analysis (FBTRCA), for pre-movement decoding.
• To enhance the accuracy of detecting upper limb movement states from EEG signals.
• To improve control capabilities for assistive technologies.

Main Methods:

• The proposed FBTRCA method incorporates filter bank selection into standard task-related component analysis (STRCA).
• EEG signals are divided into sub-bands, and STRCA is applied to extract canonical correlation patterns (CCPs).
• Minimum redundancy maximum relevance feature selection and a binary support vector machine classifier are used for classification. A convolutional neural network (CNN) was also evaluated.

Main Results:

• FBTRCA achieved an average accuracy of 0.8968 ± 0.0847 in classifying movement vs. resting states, outperforming STRCA (0.8228 ± 0.1149) and CNN (0.8828 ± 0.0917).
• For classifying two different actions, FBTRCA achieved 0.7178 ± 0.1274 accuracy, compared to STRCA (0.6611 ± 0.1432) and CNN (0.6993 ± 0.1271).
• Filter bank selection in FBTRCA demonstrated improved classification performance.

Conclusions:

• The FBTRCA method effectively improves pre-movement decoding by incorporating filter bank selection.
• This advancement offers a more natural and non-invasive control method for external devices for individuals with severe motor disabilities.
• The findings highlight the potential of optimized EEG signal processing for neuroprosthetic applications.