Extracting functional components of neural dynamics with Independent Component Analysis and inverse Current Source Density

Abstract

Local field potentials have good temporal resolution but are blurred due to the slow spatial decay of the electric field. For simultaneous recordings on regular grids one can reconstruct efficiently the current sources (CSD) using the inverse Current Source Density method (iCSD). It is possible to decompose the resultant spatiotemporal information about the current dynamics into functional components using Independent Component Analysis (ICA). We show on test data modeling recordings of evoked potentials on a grid of 4 × 5 × 7 points that meaningful results are obtained with spatial ICA decomposition of reconstructed CSD. The components obtained through decomposition of CSD are better defined and allow easier physiological interpretation than the results of similar analysis of corresponding evoked potentials in the thalamus. We show that spatiotemporal ICA decompositions can perform better for certain types of sources but it does not seem to be the case for the experimental data studied. Having found the appropriate approach to decomposing neural dynamics into functional components we use the technique to study the somatosensory evoked potentials recorded on a grid spanning a large part of the forebrain. We discuss two example components associated with the first waves of activation of the somatosensory thalamus. We show that the proposed method brings up new, more detailed information on the time and spatial location of specific activity conveyed through various parts of the somatosensory thalamus in the rat.

Generalized theory for current-source-density analysis in brain tissue

“…Similarly, corrections to the CSD given by the "forward" approach (see, for example, Ref. [21]) may also need to be reformulated for nonresistive media.…”

Independent Components of Neural Activity Carry Information on Individual Populations

“…An example can be the analysis of fMRI data [30] or of averaged stimulus-evoked potentials, where the signals are related to specific brain computation performed by a network with fixed anatomical connections [8].…”

“…Still, the obtained sources are superpositions of different overlapping populations. To extract activity of individual populations one can then use different techniques for signal decomposition, for instance independent component analysis (ICA [6], [7]) on which we shall concentrate in the present work, and indeed, success of several such approaches has been reported [8]–[10].…”

“…In the previous tests of the combinations of CSD analysis with component decompositions simple sources were typically used. For example, in [8] as the test data we used linear combinations of products of spatial sources and temporal profiles of the form (1). The coefficients were functions of time of different classes: oscillatory functions (white noise low-pass filtered under 300 Hz), simulated evoked potentials and experimental evoked potentials.…”

“…This is a method of LFP analysis we proposed in [8], the main modification being an improved method of source reconstruction (kCSD rather than iCSD). Our goal here is to use ground truth data generated with a complex network model of thalamocortical loop [14] to understand the functional meaning of components obtained from data accessible experimentally.…”