An ICA-based framework for joint analysis of cognitive scores and MEG event-related fields.

Abstract

This paper proposes an independent component analysis (ICA)-based framework for exploring associations between neural signals measured with magnetoencephalography (MEG) and non-neuroimaging data of healthy subjects. Our proposed framework contains methods for subject group identification, latent source estimation of MEG, and discriminatory source visualization. Hierarchical clustering on principal components (HCPC) is used to cluster subject groups based on cognitive scores, and ICA is performed on MEG evoked responses such that not only higher-order statistics but also sample dependence within sources is taken into account. The clustered subject labels and estimated sources are jointly analyzed to determine discriminatory sources. Finally, discriminatory sources are used to calculate global difference maps (GDMs) for the summary. Results using a new data set reveal that estimated sources are significantly correlated with cognitive measures and subject demographics. Discriminatory sources have significant correlations with variables that have not been previously used for group identification, and GDMs can effectively identify group differences.