Using the Partial Directed Coherence to Assess Functional Connectivity in Electroencephalography Data for Brain-Computer Interfaces

In this paper, we propose a statistical selection procedure by which various mental tasks can be characterized by specific brain functional connectivity. Different connectivity patterns are identified by the partial directed coherence (PDC) which is a frequency-domain metric that provides information about directionality in the interaction between signals recorded at different sensors. The basis of our selection is a statistical analysis of the directed connectivities revealed by their repeated appearance and larger PDC magnitudes in sets of electroencephalography (EEG) sensors treated as networks. Hence, our proposed method identifies significant differences between directed connectivities on EEG-sensor networks that are specific to the mental tasks involved. A combinatory analysis of different possible networks allows us to find those that characterize and discriminate the tasks and, as proof-of-concept, we analyze the connectivities of movement imageries (MIs) used in the operation of a brain-computer interface. The directed interconnections revealed by our proposed method are in agreement with brain functional connectivities already reported for MIs, and good classification rates are achieved when such interconnections are used as features in a Mahalanobis-distance-based classifier.