Space-time recurrences for functional connectivity evaluation and feature extraction in motor imagery brain-computer...

Area of Science:

• Neuroscience
• Computer Science
• Signal Processing

Background:

• Brain-computer interfaces (BCIs) rely on accurate electroencephalography (EEG) signal analysis for motor imagery classification.
• Existing functional connectivity methods face challenges in capturing complex neural interdependencies.

Purpose of the Study:

• To compare the performance of different functional connectivity frameworks for EEG-based motor imagery classification.
• To introduce and evaluate a novel recurrence-based approach for estimating functional connectivity.

Main Methods:

• Functional connectivity was estimated using Pearson correlation, Spearman correlation, mean phase coherence, and a proposed recurrence-based method.
• Graph theory metrics (clustering coefficient, degree, betweenness centrality, eigenvector centrality) were extracted.
• Fisher's discriminating ratio was used for feature selection, and a least squares classifier was employed.

Main Results:

• The recurrence-based functional connectivity estimation significantly outperformed classical similarity measures.
• No significant performance differences were found among graph features, but eigenvector centrality offered the best processing time.
• Optimal graph attributes localized to motor cortex regions correlated with subject performance.

Conclusions:

• Recurrence-based functional connectivity is a superior method for EEG-based motor imagery classification in BCIs.
• Graph-based network analysis provides valuable insights into brain functional organization.
• This study advances signal processing techniques for BCI applications.