EEG sensorimotor rhythms' variation and functional connectivity measures during motor imagery: linear relations and
Carlos A Stefano Filho1,2, Romis Attux2,3, Gabriela Castellano1,2
1Neurophysics group, "Gleb Wataghing" Institute of Physics, University of Campinas, Campinas, São Paulo, Brazil.
Peerj
|November 15, 2017
Summary
Motor imagery (MI) alters brainwave synchronization, impacting electroencephalography (EEG) signals. This study explored brain functional connectivity alongside EEG power spectral density (PSD) for brain-computer interfaces (BCI), finding PSD slightly superior for classification.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Signal Processing
Background:
- Motor imagery (MI) induces changes in neural synchronization, reflected in electroencephalography (EEG) power spectral density (PSD) of mu and beta bands.
- These EEG alterations are foundational for brain-computer interfaces (BCI), enabling command assignment.
- Emerging evidence suggests that incorporating brain functional connectivity may reveal crucial information missed by traditional PSD analysis alone.
Purpose of the Study:
- To investigate the linear correlation between MI-induced variations in EEG PSD (mu and beta bands) and alterations in brain functional networks.
- To assess the feasibility of using functional connectivity parameters as features for MI-BCI classification.
- To compare a novel graph-based functional connectivity approach with traditional PSD methods for MI-BCI performance.
Main Methods:
- The brain was modeled as a graph, with EEG electrodes as nodes.
- Functional connectivity parameters were extracted and analyzed for correlations with PSD variations in mu and beta bands during MI.
- Three feature selection techniques were explored for the graph-based method, and its classification performance was compared against the PSD method.
Main Results:
- Significant correlations (p < 0.05, 0.4-0.9) were found between PSD variations and functional network alterations, particularly in the beta band.
- The traditional PSD method yielded higher classification accuracies (90% mu, 87% beta) compared to the graph method (83% for both bands).
- The graph method required a larger number of features, though classification results were comparable when considering accuracy uncertainties.
Conclusions:
- While PSD variations correlate with functional network changes during MI, the PSD method currently offers slightly better classification performance for MI-BCI.
- The graph-based functional connectivity approach, despite its larger feature set, shows potential and comparable results, suggesting further research.
- Exploration of alternative graph metrics may enhance the efficacy of functional connectivity-based approaches for future MI-BCI development.
