Simulated data generated by the plugin can be analyzed by BrainVoyager QX for testing or didactic purposes. Major steps to analyze the generated data from the provided GDS sample file "Design_1W1B.gds" are described here; for further details of GLM and ANOVA analysis, consult the User's Guide. The figure below shows the content of a folder before running the plugin showing only a Talairach VMR file ("CG2_3DT1FL_SINC4_TAL.vmr") and the GDS file "Design_1W1B.gds". After running the Group Data Simulator with the GDS file as input, the folder contains the generated data for the requested 20 subjects. The figure below shows the data for the first five subjects:
The folder contains three files per subject, a VTC file, a protocol (PRT) file and a design matrix (RTC) file. The files of each subject can be identified by the "S
To run a GLM using the data of subject 7, enter the "Single Study General Linear Model" dialog. You will see the protocol generated for this subject (see below). As requested in the GDS file, each of the four main conditions ("Faces", "Houses", "Objects", "Patterns") occurrs three times in the protocol. Furthermore, a baseline conditon has been inserted between all main conditions as requested. To generate the design matrix, simply click the "Define Preds" button; alternatively you may also load the generated design matrix file "S7_G1_VisualExperiment.rtc".
To run the GLM for the subject's data, click the "GO" button. You will find the two modelled VOIs when moving the VMR cross through the visual cortex. The snapshot below shows VOI 1 as well as its time course after CTRL-clicking it (right time course panel). The time course of a single voxel is shown as comparison (left time course panel). The time course of a single voxel can be obtained after restricting the region growing for CTRL-clicking to value "1" in the "Max cluster spread range" field in the "Talairach" tab of the "3D Volume Tools" dialog. Note that the noise is much more visible in the single voxel time course as compared to the whole-VOI time course due to averaging of the Gaussian noise across neighboring voxels. The left side shows the "Voxel Beta Plot" from a voxel of VOI 1 following nicely the specified pattern of effects across conditions as specified for subjects of group 1 in VOI 1. It is instructive to test specific contrasts such as "Faces > Houses" using the "Overlay GLM Contrasts" dialog.
As an instructive exercise, you may want to run a multi-subject analysis separately for each group, i.e. one GLM for subjects 1 to 10 and one GLM for subjects 11 to 20. We will, however, now proceed directly run a multi-subject GLM using all subjects, which will also prepare subsequent two-factorial ANOVA analysis. Running a multi-subject RFX GLM requires creation of an appropriate multi-subject design matrix file. To create this file, enter the "General Linear Model: Multi-Study, Multi-Subject" dialog and add each subject's VTC and design matrix (RTC) file:
After adding all 20 subjects, check the "RFX GLM" option because we want to run a random effects analysis, which also prepares for a subsequent ANOVA analysis. It is also useful to save the multi-subject design matrix file (e.g. as "SimulatedVisualExperiment.mdm") for later use, e.g. ROI-GLM analysis. You may also want to change the default name for the resulting GLM file in the "Resulting GLM file" field from "model.glm" to a more useful name such as "SimulatedVisualExperiment_rfx.glm". To run the multi-subject random effects GLM, click the "GO" button. As in the single-subject case, both VOIs will be active exhibiting significant overall effects. To better see the VTC voxels for the small VOI 1, you may turn off the "Trilinear Interpolation" option in the "Overlay GLM Options" dialog. Since we have calculated a multi-subject RFX GLM, the "Voxel Beta Plot" now shows the beta values for all 20 subjects as different lines. The four conditions are shown with disks in the condition colors defined in the GDS file. To aid in visual separation of the two groups, the lines of the first 10 subjects have been set to a different color (light blue) than the 10 subjects of the second group (light yellow):
In the image above, the left panel shows the "Voxel Beta Plot" for a voxel located in the VTC but outside of the two VOIs (a "background" voxel). The middle panel shows the beta plot for a voxel in VOI 1 and the right panel shows the betas for all subjects in VOI 2. Note that with 10 subjects, the overall effect (mean of all conditions vs baseline) is significant at both VOIs even with multiple comparison correction (FDR and Bonferrroni). If, however, the contrast "Faces > Houses" is overlaid, only the first VOI reaches FDR-based significance but not VOI 2. This is explained if one considers that the effects of all subjects are pooled in this GLM analysis, which does not reveal the strong effects in each subgroup, which is visible as an interaction effect in the voxel beta plot for VOI 2 (first two columns in right panel above). This can be also shown by running VOI RFX GLM analyses for both VOIs, which can be defined by overlaying the overall effects contrast. To reveal that both groups differ in the "Faces > Houses" contrast, a random effects group comparison can be performed by using the "summary statistics" approach. In this approach, the subjects are categorized into the two groups. For each subject the difference of the "Faces" and "Houses" betas is calculated as the summary statistic and a t test is performed comparing the means of this difference across the two groups. This "old style" (but valid) analysis is availble in the "Random effects analysis (old version)" field in the "Overlay GLM Options" dialog.
A more flexible random effects analysis approach is provided by the "ANCOVA" dialog allowing to spcify multi-factorial designs with both within and between factors. The multi-factorial analysis requires a RFX-GLM (or standard separate-subject GLM) as input providing estimates of effects (beta values) for each subject. In order to run an ANOVA analysis for the generated data, open the "ANCOVA" dialog. If you ocntinue right away from the computed RFX-GLM, the beta values are already available and the design can be specified. If no appropriate GLM data structure is available, browse to a GLM file (e.g. "SimulatedVisualExperiment_rfx.glm") in the "GLM / AVA" tab of the "ANCOVA" dialog. The factorial design is specified in the "Design" tab, which shows one within factor as default listing all conditions found in the provided GLM including a "Baseline" condition. Since the within conditions are already listed for the within factor, you only may want to change its name, e.g. to "Visual Stimuli". Since we also want to model a between factor, change the "Nr of between factors" spin box from "0" to "1". To define the levels of the between factor, click on the "Between 1" entry in the "Factor list". This updates respective fields showing the name of the factor in the "Factor name" field, the number of levels of the factor in the "NrOfLevels" spin box, and the names of the factor levels in the "Level list".
Change the name of the selected (current) between factor to "Sex". The number of levels (default "2") of this factor needs not be changed since the design in the sample GDS file specifies also 2 levels. Change the default level names ("Level 1", "Level 2") to the names used in the GDS file ("Male", "Sex"). Switch to the "Table" tab, which shows a table with data for each subject. This table is filled if the dialog is used in the context of a VOI ANCOVA analysis. When a statistical map is computed as in our case, the table is filled with "0" values. The table contains, however, one important piece of information, namely how the subjects are grouped with respect to the between factor(s). This information is provided in the column on the right side, which contains the name of the between factor ("Sex") in the table header. All subjects are as default assigned to level 1 (group "Male"). To specify that subjects 11 - 20 belong to group 2 ("Female"), change the value from "1" to "2" for the rows of these subjects. You may save the specified design by clicking the "Save Design" button (e.g. as "SimulatedVisualExperiment.ads"). To run the 2-factorial ANOVA analysis, click the "GO" button.
The two-factorial ANOVA model allows for testing the significance of overall main effects for each factor as well as for significant interaction between the two factors. To specify one of these tests, check the respective option in the "Overlay RFX ANCOVA Tests" dialog and click "OK", which will result in a corresponding F map. It is also possible to test specific contrasts by checking the "Specify contrast" option and by editing the values in the provided contrast table (see snapshots above and below). The snapshot above shows how the same contrast ("Faces > Houses") can be formulated across groups. The resulting t map shows a strong effect in VOI 1 but not in VOI 2, as expected from the beta plots (see above). The snapshot below shows how a specific interaction effect across groups can be tested, e.g. is the difference "Faces - Houses" different for the "Male" vs the "Female" group?. The resulting t map now shows no significant effect for VOI 1 but a highly significant effect for VOI 2 as one would predict from the beta plot shown above.
Besides calculating maps, it is possible to run the ANOVA analysis for any region-of-interest providing detailled numerical output. As a prerequisite, VOIs have to be defined from functional clusters or anatomically. Define the two simulated VOIs by overlayng an overall effects contrast, CTRL-clicking the two functional clusters, right-clicking the shown time course window and clicking the "Define VOI" button in the appearing context menu. After defining the two VOIs, invoke the "ANCOVA" dialog from the "VOI Analysis Options" dialog, which itself can be called from the "Volume-Of-Interest Analysis" dialog. In the "Design matrix file" text box of the "GLM Options" tab, specify the same multi-subject design matrix file ("SimulatedVisualExperiment.mdm") as used to calculate the RFX-GLM above. When running the ANOVA, the design matrix file is used to run a RFX-GLM over the VOI time course of all subjects in order to obtain VOI beta values. To launch the ANCOVA dialog, switch to the "VOI GLM" tab and click the "ANCOVA" button in the "VOI RFX analysis" field. You will see the "Design" and "Table" tabs in the "ANCOVA" dialog. The table in the "Table" tab contains the estimated VOI beta values for each subject. Switch to the "Design" tab, which shows the default settings for a one within factor design. To add the second factor of the simulated design, you may enter the second factor ("Sex") as described above. A more convenient solution is to reload the previously saved design specification. To reload the stored design, click the "Load Design" button and open the previously saved design file ("SimulatedVisualExperiment.ads"). If the program asks you whether to load the values in the design file (table data), select "No", otherwise the calculated RFX GLM beta values would be overriden with the values in the design file (which are all "0"). Switch to the "Table" tab. The right column contains the group factor "Sex" with the two levels "1" ("male") and "2" ("female"). To specify that the subjects 11 - 20 belong to group 2, change the respective entries in the "Sex" column:
