Multivariate Empirical Mode Decomposition of EEG for Mental State Detection at Localized Brain Lobes

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Electroencephalography (EEG) is crucial for monitoring brain activity.
• Accurate mental state classification from EEG is challenging due to signal complexity.
• Existing methods like Fourier and Wavelet transforms have limitations in capturing non-linear EEG dynamics.

Purpose of the Study:

• To apply the Multivariate Empirical Mode Decomposition (MEMD) approach for enhanced feature extraction from multi-channel EEG signals.
• To evaluate the efficacy of MEMD-derived features for mental state classification.
• To identify dominant brain regions contributing to mental state detection using MEMD.

Main Methods:

• Utilized Multivariate Empirical Mode Decomposition (MEMD) to decompose multi-channel EEG signals into intrinsic mode functions (IMFs).
• Extracted non-linear features from high-oscillation IMFs identified as most relevant for mental state discrimination.
• Compared MEMD features against conventional tempo-spectral features from Fourier and Wavelet transforms.
• Analyzed MEMD features from specific EEG channels to determine regional brain dominance.

Main Results:

• MEMD features demonstrated significant performance gains over traditional tempo-spectral features.
• High-oscillation IMFs were found to be most informative for distinguishing mental states.
• Classification accuracy reached 98.06%, with the frontal brain region showing the most significance.
• The MEMD approach effectively captures joint channel properties for discriminative feature extraction.

Conclusions:

• MEMD is a powerful, data-adaptive tool for analyzing complex, multi-dimensional EEG signals.
• MEMD-based feature extraction offers superior performance for EEG-based mental state detection.
• The frontal region plays a critical role in mental state classification, as highlighted by MEMD analysis.