Reproducibility analysis of functional connectivity measures for application in motor imagery BCIs

Electroencephalography (EEG)-based motor imagery brain-computer interfaces (MI-BCIs) can improve motor rehabilitation processes. Nevertheless, the large variability of intra and inter-subject EEG signals has precluded translation of this technology to the clinical setting. The aim of this work was to evaluate the reproducibility of EEG functional connectivity (FC) features used to discriminate between left-and right-hand MI. Ten subjects underwent 12 EEG-MI-BCI sessions. Two frequency bands, three FC methods, four graph parameters and six electrode sites were evaluated, using statistical and classification analyses. The & beta; band produced the largest number of statistically significantly discriminating and most stable features for the majority of subjects, and also the best classification features, suggesting that engagement of needed brain regions may be more important for stability and distinguishability among MI tasks than inhibition of unneeded cortical regions. The Cz electrode stood out in terms of largest number of statistically significantly discriminating features. The motif synchroni-zation (MS) method produced the largest number of significantly discriminating features, the most stable, most discriminating features and best classification features for most subjects. Since this method ignores amplitude changes, this seems to indicate that signal variation patterns are more important for feature stability and class separability. Eigenvector centrality (EC) was the most stable graph parameter for most subjects while both EC and strength (S) were the most discriminating. In summary, using features from the MS method, & beta; band, EC and/ or S parameters, and central electrodes (Cz), result in the best combination for the task of distinguishing left from right-hand MI.