Mapping multi-modal dynamic network activity during naturalistic music listening

Abstract

Music is an intricate stimulus that engages numerous brain networks at various temporal scales. Past work has successfully correlated brain regions and networks of brain regions with features in musical signals and behavioural ratings. Current dynamic systems frameworks allow for the incorporation of multiple types of high-dimensional stimuli into the same model. In the present study, we used networks of music features and continuous behavioural ratings to examine patterns of brain network connectivity captured via EEG in two music listening conditions. Participants listened to music while continuously rating either perceptual features in the musical signal or emotional states induced by the music. Networks of brain regions and networks of music features were estimated using hidden Markov modelling and were analyzed together with continuous behavioural ratings using partial least squares analysis. We found task-specific differences in between-network brain activity. In the perceptual task, network activity in medial frontoparietal, pericentral, anterior frontal, and bilateral temporal networks was most strongly positively correlated with acoustic features in the musical signal. In the emotional task, transitions between networks were more positively correlated with low-arousal low-valence emotional states in anterior cingulate, frontal, and occipital networks. We conclude that attention to specific aspects of a musical experience employs different network connectivity patterns in music listening and discuss the implications of these methods for future work on feature-rich naturalistic behaviours.

Dynamic network features of functional and structural brain networks support visual working memory in aging adults

“…HMM was initially developed for applications to protein and DNA sequencing, and has been used widely for modelling data that follows a sequence, while LEiDA, developed for neuroimaging data, accomplishes similar goals while integrating a step that extracts the leading eigenvector from time × time phase coherence matrices, capturing dominant connectivity patterns while diminishing the effects of noise. Both of these approaches have been applied to neuroimaging data revealing distinct brain states at different points in time relying on separable combinations of regions (e.g., Cabral et al, 2017;S. E. Faber et al, 2024; S. E. M. Faber et al, 2023;Vidaurre et al, 2017).…”