Real-Time Independent Component Analysis
Independent component analysis (ICA) can identify brain activity from functional magnetic resonance imaging (fMRI) time-series without a priori "temporal" assumptions, i e. with no detailed information about the experimental
design or the expected hemodynamic response shape and timing, and with only rough knowledge of potentially activated areas in relation to stimulus processing or brain status changes.
Using a sliding-window approach for real time data preparation, a spatial ICA analysis can be conveniently restricted to the most recently acquired data. Assuming that both edges of the window move during the acquisition, both the accuracy and the computational load of the real-time analysis is constant over time, whereas the sensitivity to dynamic changes in brain activity is maximized.
Real-time fMRI enables one to monitor a subject's brain activities during an ongoing session, but results are to be delivered within times in the order of one or a few TR(s). In order to deliver ICA components as fast as possible the deflation scheme of the FastICA algorithm by Aapo Hyvarinen is exploited. This makes the ICA maps immediately available for selection and display by the user even if not all possible components have been extracted from the full dimensionality window data sets. Moreover, it has been demonstrated that the sliding-window FastICA algorithm can make a real-time ICA analysis perform comparably to a general linear model (GLM) analysis, without the need of critical settings for the algorithm, provide that the user is focused on relatively few consecutive slices in the real-time acquisition.
The principal motivation for running a real-time ICA analysis with the real-time ICA plugin is that no timing information (e. g. a protocol) is to be specified by the user before starting a TBV session. This opens the possibility of monitoring a non-triggered, non-repetitive and non-stationary neural activity with only minimal spatial prior on the networks involved. The spatial prior is easily and dynamically defined and updated over time by the user in terms of rectangular masks including the potential regions of interest (ROI). Moreover, integration of rt-ICA generated maps in neurofeedback experiments is possible since the plugin can optionally save the component data to disk.
The next topic describes how to use the real-time ICA plugin.
Esposito F, et al. (2003). Real-time independent component analysis of fMRI time-series. Neuroimage, 20, 2209-2224.
Hyvarinen A. (1999). Fast and robust fixed-point algorithms for independent component analysis. IEEE Trans Neural Netw, 10, 626-634.
Copyright © 2014 Rainer Goebel. All rights reserved.