Comparison of filtering and classification techniques of electroencephalography for brain-computer interface

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalography (EEG) is crucial for understanding brain activity during motor tasks.
• Extracting meaningful features from EEG signals, particularly mu rhythms, is challenging.
• Accurate classification of EEG features is essential for brain-computer interfaces and clinical applications.

Purpose of the Study:

• To investigate and compare various feature extraction and classification methods for EEG mu rhythms.
• To evaluate the effectiveness of autoregressive (AR) filtering, mu-matched filtering, and wavelet decomposition (WD) for feature extraction.
• To assess the performance of linear classifiers and support vector machines (SVMs) in classifying extracted EEG features.

Main Methods:

• Feature extraction using autoregressive (AR) filtering, mu-matched filtering, and wavelet decomposition (WD).
• Classification of extracted EEG features using a linear classifier with expert-defined weights.
• Classification using support vector machines (SVMs) with diverse kernel functions.
• Comparative analysis of classification accuracies across different methods.

Main Results:

• Support vector machines (SVMs) demonstrated potential for improved classification accuracy compared to the simple linear classifier.
• Wavelet decomposition (WD) and mu-matched filtering showed potential advantages over autoregressive (AR) filtering for feature extraction.
• The study provides a quantitative comparison of the investigated feature extraction and classification techniques.

Conclusions:

• SVMs offer a promising advancement for EEG feature classification in motor task analysis.
• Wavelet decomposition and mu-matched filtering are effective alternatives for EEG feature extraction, outperforming AR filtering.
• These findings contribute to the development of more accurate and robust EEG-based analysis methods.