Event Abstract Back to Event Spatiotemporal decomposition of four-dimensional MEG beamformer source reconstructions Paul Ferrari1*, Marc Lalancette1 and Douglas Cheyne2 1 The Hospital for Sick Children, Program in Neurosciences and Mental Health, Canada 2 University of Toronto, Program in Neurosciences and Mental Health, Hospital for Sick Children, and Department of Medical Imaging, Canada Developments in MEG source reconstruction techniques have resulted in the ability to estimate brain activity in multiple dimensions with high spatial, temporal, and spectral resolutions[1]. These advances come with the added difficulty associated with vast increases in data dimensionality. This includes both the statistical detection of low amplitude source activity in the presence stronger sensory or motor responses, and the characterization of large-scale network activation embedded within this high dimensional source space. We describe the application of principal components analysis (PCA) decomposition for characterizing spatiotemporal activity within both simulated datasets of known source composition and experimental data from subjects performing sensorimotor tasks. PCA was performed by eigenvalue decomposition of the covariance matrix of the four-dimensional beamformer source space and volumetric patterns contributing to a majority of the variance were found by projecting the beamformer data separately onto the highest ranking eigenvectors. This decomposition results in unique spatial modes that (1) are common to all subjects (2) are ranked by their contribution to the variance, and (3) have distinct time courses. PCA extracted all source activity in the simulated data, despite overlapping source generators of high and low signal to noise. Analysis of the experimental data showed that the 1st spatial mode, as expected, identified contralateral sensorimotor cortex. Investigation of higher components revealed weaker activation patterns in ant. cingulate, premotor, ipsilateral sensorimotor, and parietal cortex. These results demonstrate the feasibility of network reconstruction from MEG beamformer data via spatiotemporal decomposition techniques. Extension of this work will include independent components decomposition and the functional categorization of different component structures.