Analyzing Real and Simulated Group Data
One of the new features of BrainVoyager QX 1.8 is the “GroupDataSimulator” plugin allowing to easily generate simulated multi-subject data for various experimental designs. For as many subjects as desired, the simulator generates functional data (VTC files), protocol and design matrix files. Why would one want to generate artificial data?
Working with simulated data is very useful for testing multi-subject analysis tools since one knows for sure where specific activation effects are located “in the brain”. By controlling the amount of noise in the simulated data, one also knows how difficult it is for hypothesis-driven and data-driven tools to detect the “hidden” activation patterns. Generating data with “ideal” activation patterns may also be helpful to investigate the number of subjects needed to obtain significant random effects results with different methods under various conditions going beyond standard power analyses. It might also be instructive to compare results obtained with simulated data with the results from measured real data using the same analysis steps and visualization tools. Furthermore, simulation of complex experimental data may play an important didactic role: If one is able to successfully analyze simulated data of a planned experiment, one should have learned everything necessary to successfully analyze also subsequently measured real data.
I originally created the plugin to be able to extensively test the code implementing the new multi-factorial designs available in the enhanced ANCOVA module of QX 1.8. This module can be used to perform random effects (RFX) analyses for multi-factorial designs (second level analysis). The dependent variable are beta values for experimental conditions estimated with a separate-subject GLM (first level analysis). The updated module adds several important multi-factorial designs, which were missing in earlier versions. The previous version only allowed to model repeated measures (within factors) designs but did not allow to calculate maps for designs with between (grouping) factors. A design with one between factor and one within factor adequately models many common experiments comparing fMRI effects across various experimental groups. Examples for between factors are “sex” (with levels “male” vs “female”), “reading ability” with levels “dyslexic” vs “non-dyslexic”, or “disease” with levels “disease 1”, “disease 2”, “disease 3” and so on. The number of levels of between and within factors is unlimited. The plot above shows estimated beta values of a voxel for 20 subjects with four experimental conditions of a within factor and two groups of a between factor. Subjects belonging to group 1 are depicted with blue lines, subjects belonging to group 2 are depicted with light blue lines. These beta values constitute the data analyzed by a two-factorial ANOVA. To calculate brain maps, such a two-factorial ANOVA analysis is performed independently for each voxel.
Main effects of such a 1-between, 1-within design can be tested using F tests, which can be selected in the “Overlay RFX ANCOVA Tests” dialog. For a between factor “sex”, the calculated F map would, for example, show brain regions with significantly different effects for “male” and “female” subjects across the conditions of the within factor. The F map for the within factor (e.g. repeated measures factor “visual stimulation” with three levels “faces”, “houses”, “cars”) would show brain regions with significantly different responses of “visual stimulation” conditions pooled across the two “sex” groups. A third F test can be selected to test for an interaction effect between the two factors revealing brain regions showing non-additive effects. Specific contrasts between factor cells (e.g. “male-cars > female-cars”can also be tested resulting in respective t maps. For an example analysis of such a design, consult the online help of the “GroupDataSimulator” plugin. For details about GLM and ANOVA analysis, consult the BrainVoyager QX User’s Guide.
The release notes of BrainVoyager QX 1.8 describe further new features, enhancements and bug fixes. The program is currently running through a newly established quality assurance procedure and should be available for download on the BrainVoyager website in a few days. Work is turning to BrainVoyager QX 1.9 now, which is expected to contain a nice and powerful Diffusion Tensor Imaging (DTI) module.