Human Brain Mapping 2015 Honolulu, Hawaii
June 14-18, 2015
Scott Makeig , Swartz Center for Computational Neuroscience, Institute for Neural Computation, University of California San Diego, La Jolla, CA
Blind Source Separation of Electrophysiological Data
For many decades the most widely used statistical methods were those that attempt to characterize the shape of a multivariate or multichannel data set as a Gaussian cloud, as these statistics are mathematically simple and computationally efficient. The basic data model was a single fixed value per condition plus noise whose size and shape need be characterized to estimate confidence in the sample mean and in mean value differences between conditions. However, brain electrophysiological data do not naturally fit this model. Rather, EEG (like MEG, EMG, ECoG, etc.) data channels sum potentials produced by a considerable number of non-Gaussian brain and non-brain electrical processes. EEG and MEG effective source signals can be largely categorized as field potentials from compact cortical areas in which the net electrical field signals are locally coherent, as well as non-brain processes including eye movements, line noise, scalp muscle EMG, channel movement noise, etc..
During the 1990s, the field of blind source separation revolutionized statistics by exploiting the simple and physiologically plausible assumption that the signals from such processes are temporally independent or near-independent. This allows ‘unmixing’ the linear mixtures of these signals delivered by the recording electrodes, while remaining ‘blind’ to (i.e., without foreknowledge of) the natures of the source processes. Applied to dense-array electrophysiological data, independent component analysis (ICA) identifies the source-level signals of the effective data sources as well as the projection pattern of each source to the electrode array (the component scalp or electrode grid map).
ICA source-resolved electrophysiological imaging methods are now contributing strongly to an ongoing and radical shift in human brain electrophysiology research -- from 2-D channel signal analysis to fully 3-D, highly msec- and cm2-resolved functional source imaging of distributed cortical dynamics. I will show examples computed using freely available open source tools.