Human Brain Mapping 2002
June 10-16, 2002
J-R Duann, Tzyy-Ping Jung, Scott Makeig & Terrence J. Sejnowski,
Institute for Neural Computation, UCSD & The Salk Institute, La Jolla CA
fMRLAB: An ICA Toolbox for fMRI Data Analysis
Introduction: Analysis of functional magnetic resonance imaging (fMRI) brain data is a challenging enterprise, as the fMRI signals exhibit highly varying, unpredictable time courses that represent the summation of hemodynamic influences from neural activity, subject movements and machine noise, as well as from physiological cardiac, respiratory and other pulsations. The relative contribution and exact form of each of these components in each voxel across a given session is unknown to the experimenter in advance of the analysis, suggesting a role for data-driven methods when the data are consistent with model assumptions. Independent Component Analysis (ICA), a technique for blind source separation, can separate source components that are mixed additively in the observed data. The resulting source components may be directly, indirectly, or not at all related to a particular performance time course. ICA has been shown to be a powerful tool for exploratory fMRI analysis (McKeown et al., Human Brain Mapping, 1998). However, exploring the scientific interest of the resulting ICA components can appear an unmanageable task. There is a need, therefore, for software tools to streamline ICA component derivation and evaluation.
Methods: Here we describe a GUI-based software toolbox for fMRI data analysis based on Infomax ICA using the MATLAB signal processing environment (The Mathworks, Inc.). fMRLAB (freely available online at sccn.ucsd.edu/fmrlab/) provides functions for data preprocessing, ICA source separation, and component review for fMRI time series analysis. Preprocessing currently includes slice-timing adjustment and off-brain voxel removal. Post-processing functions include: (1) defining and visualizing the Region Of Activity (ROA) of each component, (2) computing and visualizing the time course of each component in the data, compared to the mean time course of positive-valued voxels in the component ROA, (3) converting the calculated time courses to percent signal change, (4) assessing the percentage variance accounted for by each component within the ROA mean, and (5) finding and visualizing the components accounting for the most variance in the ROA-mean signal.
Results: fMRLAB allows researchers to readily decompose fMRI signals into maximally spatially independent components, then to rapidly search through the resulting components, noting for each component its most active brain areas and its associated hemodynamic time course. BOLD-image plots (Duann et al., HBM'01) visualize the quality of time-locking of the unaveraged component time courses to experimental events. Components of interest can be noted and recalled. Finally, the data structure containing fMRLAB results can be stored on disk and/or called from the MATLAB commandline for further analysis.
Conclusion: Blind source separation of unaveraged and unregressed functional imaging data allows researchers to determine and evaluate relationships between ongoing or event-related changes in hemodynamics and behavior and to perform exploratory and hypothesis-driven analyses on human brain data for basic and clinical studies of human brain function. Using fMRLAB streamlines the ICA analysis and results evaluation process.
Return to Current Abstracts