Robust classification of motor imagery EEG signals using statistical time-domain features

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Real-time brain-machine interface (BMI) systems require computationally efficient methods.
• Traditional time-domain features like Willison amplitude (WAMP) and slope sign change (SSC) depend on optimal threshold selection.
• Trial-and-error threshold determination is a significant drawback for WAMP and SSC.

Purpose of the Study:

• To introduce modified WAMP and modified SSC to eliminate the need for trial-and-error threshold selection.
• To comprehensively assess statistical time-domain features for BMI applications.
• To evaluate feature-classifier combinations using Support Vector Machine (SVM) and supervised fuzzy C-means.

Main Methods:

• Development of modified Willison Amplitude (WAMP) and modified Slope Sign Change (SSC) algorithms.
• Evaluation of various statistical time-domain features using Support Vector Machine (SVM) classifier.
• Cross-validation of SVM results with supervised fuzzy C-means for enhanced accuracy assessment.

Main Results:

• Modified WAMP and SSC effectively address the threshold selection challenge.
• Most subjects achieved performance levels near or exceeding 80% with the evaluated features.
• Feature combinations significantly improved classification accuracy, reaching up to 100% in some cases.

Conclusions:

• The proposed modified features and feature-classifier combinations are suitable for real-time BMI applications where minor errors are tolerable.
• Selected features, when combined, demonstrate a substantial enhancement in BMI performance.
• This research offers efficient and accurate solutions for advancing real-time brain-machine interface technology.