Data acquired by functional brain imaging are of a multivariate and complex nature. Selecting relevant topographically specific information for system-level analysis is a highly non-trivial task. This challenge has traditionally been addressed by hypothesis-driven approaches. Recently, data-driven methods making no a priori assumptions about the signal were developed. Here, we present a hybrid approach, selecting data-driven voxels in paradigm-driven measurements in order to identify functional connectivity motifs in the voxel correlations. Our tool is the functional holography (FH) method, originally developed for analyzing electrophysiological recordings and based on analyzing the voxel-voxel correlation matrices. The algorithm selects the relevant voxels using a dendrogram clustering method combined with a unique standard deviation (STD) filter, identifying the voxels with high STD correlations. Functional connectivity motifs are revealed through a dimension-reduction procedure by principal component analysis (PCA) allowing for a reduced three-dimensional holographic presentation space. Information loss due to PCA is retrieved by connecting voxels in the reduced space with lines that are color-coded according to the correlations. Our results show that the FH analysis performed for a single trial reveals interesting motifs, even in a simple motor task: unilateral hand movements yielded two clusters, one in the contralateral M1 region showing neuronal activation and one in the ipsilateral homologues region showing deactivation. Thus, according to a single trial level analysis, of 12-time points alone, we can determine which hand the subject moved. Moreover, using cluster quantification based on eigenvalue entropy calculation, we obtained good separation between right- and left-handed subjects.
(c) 2010 Elsevier B.V. All rights reserved.