Graph Signal Processing as a tool for mitigating the impact of spatial blurring in EEG-based neuroelectrical imaging

Spatial blurring phenomena in EEG scalp maps are strongly influenced by the topology of the acquisition domain. Together with the volume conduction effect, the positioning of the electrodes on the scalp is a key factor affecting the spatial resolution of EEG-based scalp maps. In this perspective, modern Graph Signal Processing (GSP) techniques can be used to decompose the EEG signal into network harmonics to unveil the contribution of different spatial components on the original signal. In this regard, it is reasonable to assume that crosstalk phenomena among adjacent electrodes can be expressed as a linear combination of slow varying network harmonics. A tailored graph filtering procedure may thus be helpful in mitigating the spatial blurring induced by crosstalk phenomena in EEG scalp maps. In line with this, we investigated how different graph filtering procedures affect the spatial resolution of grand average scalp maps extracted from a group of 15 healthy subjects involved in the execution of simple motor tasks. Results showed that scalp localization can be improved using a high-pass graph filtered version of the EEG signal. On the contrary, the contribution of slow-varying network harmonics describes a scalp pattern that lacks spatial localization and contributes to blur the scalp maps.