Toward automatic detection of brain responses to emotional music through analysis of EEG effective connectivity

Abstract

The purpose of this study was to investigate the effective brain networks associated with joyful, melancholic, and neutral music. Connectivity patterns among EEG electrodes in different frequency bands were extracted by multivariate autoregressive modeling while 19 nonmusicians listened to selected classical and Iranian musical excerpts. Musical selections were categorized according to the participants' average self-assessment results. Connectivity matrices were analyzed to identify distinct variations in the connectivity indices related to the categorized excerpts. We studied the correlation of inter-/intra-regional connectivity patterns with the self-reported evaluations of the musical selections. The perceived valence was positively correlated with the frontal inter-hemispheric flow, but negatively correlated with the parietal bilateral connectivity. Using the connectivity indices between different cortical areas and a support vector machine, we sought to distinguish trials in terms of the self-reported valence of perceived emotions and the familiarity of the musical genres. For 16 participants, the average classification accuracies in discriminating joyful from neutral, joyful from melancholic and familiar from unfamiliar trials were 93.7% ± 1.06%, 80.43% ± 1.74%, and 83.04% ± 1.47, respectively. Integration of different cortical areas is required for music perception and emotional processing. Thus, by studying the connectivity of brain regions, we may be able to develop a noninvasive assessment tool for investigating musical emotions.