Role of Scalp EEG Brain Connectivity in Motor Imagery Decoding for BCI Applications

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Multichannel electroencephalography (EEG) brain connectivity (BC) features are explored for motor imagery (MI) decoding in brain-computer interfaces (BCIs).
• The comparative advantage of BC features over single-channel EEG features for MI decoding remains unclear.
• Understanding the relationship between central nodes in BC networks and influential EEG channels is crucial for optimizing BCI performance.

Purpose of the Study:

• To investigate the relationship between central nodes in BC networks and influential single-channel EEG features for MI decoding.
• To compare the decoding accuracy of BC features in source versus sensor space.
• To evaluate the performance of different BC features (Phase Locking Value, Granger Causality, weighted Phase Lag Index) for MI classification.

Main Methods:

• Utilized three BC features: Phase Locking Value (PLV), Granger Causality, and weighted Phase Lag Index.
• Investigated the centrality of nodes in BC networks and their overlap with influential EEG channels identified via common spatial pattern (CSP) filtering.
• Compared MI decoding accuracy using BC features in sensor space versus source space on the BCI Competition VI Dataset 2a (left- vs. right-hand MI).

Main Results:

• Phase Locking Value (PLV) in sensor space yielded the highest classification accuracy among the BC features evaluated.
• PLV in sensor space demonstrated performance comparable to single-channel EEG features.
• Transitioning BC features from sensor to source space resulted in a reduction in average classification accuracy.
• Network topology analysis revealed similar patterns for left- vs. right-hand MI tasks across BC measures.
• Central nodes in BC networks showed partial overlap with channels identified as influential in single-channel classification.

Conclusions:

• Brain connectivity features, particularly PLV in sensor space, are effective for motor imagery decoding and perform comparably to traditional single-channel methods.
• Employing source space analysis for BC features may not enhance, and could potentially decrease, MI decoding accuracy compared to sensor space.
• The overlap between central BC network nodes and influential EEG channels suggests potential for integrated feature selection strategies in BCI development.