Cortical representations of Auditory Perception using Graph Independent Component on EEG

TL;DR

This study uses Graph-Independent Component Analysis on EEG data to identify independent brain subnetworks involved in music perception, revealing specific regions activated during auditory processing.

Contribution

It introduces a novel application of Graph-ICA to decompose whole-brain EEG networks into independent subnetworks related to auditory perception.

Findings

Identified subnetworks in temporal lobes and Broca's area involved in music perception.

Electrode locations correspond to known auditory and language regions.

Subnetwork activity increases with music tempo.

Abstract

Recent studies indicate that the neurons involved in a cognitive task aren't locally limited but span out to multiple human brain regions. We obtain network components and their locations for the task of listening to music. The recorded EEG data is modeled as a graph, and it is assumed that the overall activity is a contribution of several independent subnetworks. To identify these intrinsic cognitive subnetworks corresponding to music perception, we propose to decompose the whole brain graph-network into multiple subnetworks. We perform this decomposition to a group of brain networks by performing Graph-Independent Component Analysis. Graph-ICA is a variant of ICA that decomposes the measured graph into independent source graphs. Having obtained independent subnetworks, we calculate the electrode positions by computing the local maxima of these subnetwork matrices. We observe that the computed electrodes' location corresponds to the temporal lobes and the Broca's area, which are indeed involved in the task of auditory processing and perception. The computed electrodes also span the brain's frontal lobe, which is involved in attention and generating a stimulus-evoked response. The weight of the subnetwork that corresponds to the aforementioned brain regions increases with the increase in the music recording's tempo. The results suggest that whole-brain networks can be decomposed into independent subnetworks and analyze cognitive responses to music stimulus.