Correlation-Based Adaptive Filtering Performance Driven by Signal Decomposition Methods for EEG Signals of Individuals with Severe Motor Disabilities

Electroencephalogram (EEG) signals are obtained from the surface of the scalp through a Brain-Computer Interface (BCI) and show the activity of brain regions. EEG signals are useful tools for people, especially those who have severe motor disabilities for an improved quality of life. Because of their noisy nature caused by a small movement of the head, eye blink, or even breathing, it is extremely hard to extract meaningful information from EEG signals. Thus, it is imperative to filter EEG signals without losing their essential parts. While filtering, the concept of adaptivity comes in handy because of its flexible nature to preserve important information. In this context, a novel approach that applies Wavelet Decomposition (WD), Empirical Mode Decomposition (EMD), or Variational Mode Decomposition (VMD) methods to filter EEG signals adaptively based on correlation was proposed. The performances of developed methods were calculated based on the accuracies of binary classifications of five different labels by using machine learning classifiers and compared with an Elliptic Bandpass Filter. The study shows that the proposed adaptively implemented VMD is the most useful filtering method for EEG signals, providing the best performance with 76.1% subject-wise accuracy for binary classification of word association and feet motor imagery and an average accuracy of 70.4% for all binary classifications using a Support Vector Machine (SVM) classifier. The results show that correlation is an efficient tool to adaptively implement signal decomposition methods as filters by preserving meaningful information more successfully than an Elliptic Bandpass Filter.