A Frequency-Shifting Variational Mode Decomposition-Based Approach to MI-EEG Signal Classification for BCIs

Area of Science:

• Neuroscience and Biomedical Engineering
• Signal Processing

Background:

• Electroencephalogram (EEG) signal analysis is vital for neurological diagnostics.
• Traditional signal decomposition (SD) methods suffer from mode mixing and aliasing, degrading signal integrity and diagnostic accuracy.
• These limitations impact the diagnosis of conditions like epilepsy, brain injuries, and sleep disorders.

Purpose of the Study:

• To introduce a novel Frequency-Shifting Variational Mode Decomposition (FS-VMD) method for improved EEG signal analysis.
• To address and overcome the critical issues of mode mixing and mode aliasing in EEG decomposition.
• To enhance the accuracy and efficiency of EEG-based diagnostics.

Main Methods:

• Developed the Frequency-Shifting Variational Mode Decomposition (FS-VMD) technique.
• FS-VMD extracts and shifts the fundamental frequency of EEG signals to a lower range for iterative decomposition.
• Integrated FS-VMD with advanced classifiers: Support Vector Machines (SVM), Convolutional Neural Networks (CNN), and Feature-Weighted k-Nearest Neighbors (FWKNN).

Main Results:

• FS-VMD effectively reduces mode mixing and mode aliasing, enhancing intrinsic mode function (IMF) resolution.
• Achieved superior classification accuracy, with SVM reaching 99.99% in an 18-channel EEG setup (0.25 standard deviation).
• Demonstrated significant improvements in signal clarity and decomposition precision compared to traditional SD techniques.

Conclusions:

• FS-VMD offers a robust and precise solution for EEG signal analysis, overcoming limitations of conventional methods.
• The proposed method significantly enhances diagnostic accuracy for neurological conditions.
• FS-VMD represents a substantial advancement in EEG signal processing for clinical applications.