Graph Fourier Transform to mitigate the effects of crosstalk in hdEEG recordings

Electroencephalography (EEG) is probably the most popular non-invasive technique for the acquisition of brain signals. Despite its ease of use and moderate costs, this technique intrinsically suffers from poor spatial resolution, which is known to be caused by the combinations of the volume conduction effect and crosstalk phenomena. While the first one is caused by the propagation of a source signal through different biological tissues, the second one relates to the positioning of the electrodes on the scalp, manifesting as a spurious electric signal involving a set of neighboring electrodes even in absence of true brain activity. The presence of such a spurious signal not only contributes to the spatial blurring of EEG-based scalp maps, but is also known to alter brain connectivity estimate. In this work, the simultaneous engagement of adjacent electrodes typical of crosstalk was used to characterize this phenomenon in terms of network harmonics with respect to the graph structure describing the positioning of the electrodes on the scalp. In this perspective, a tailored graph filter could be used to mitigate the effects of crosstalk and improve the accuracy of multivariate brain connectivity maps. As to do so, this work investigates the effects of different graph filtering procedures on two different datasets: a set of EEG-like data recorded on a polystyrene mannequin (representing a null-case scenario for causal connectivity) and a real EEG dataset recorded from a healthy subject during the execution of simple hand movements.