Chapter 4 Example box: Visualizing dual regression group analysis results


The aim of this example is to help improve your understanding of dual regression and to get experience looking at some results.

This example is based on tools available in FSL, and the file names and instructions are specific to FSL. However, similar analyses can be performed in other neuroimaging software packages.

Please download the dataset for this example here:

Data download

The dataset you downloaded contains an example output folder after running a dual regression analysis in FSL, including a group-level comparison. In the example below, we will go through the files that exist in the output directory, and discuss how to find and visualise significant results.

Explanation of dual regression outputs

Open a command line terminal and change directory into the dual regression directory you have downloaded called 'Data_4.3' (using cd). You can now list the contents of the directory using ls for example type ls groupICA15.dr. The directory contains the following files:

  1. dr_stage1_subject**.txt: There is one text file per subject, which contain the timeseries estimated in stage one of the dual regression. Each text file has as many columns as there are maps in the group-ICA input, and as many rows as there are timepoints in the original subject data. These timeseries can also be used for node-based analyses as described in chapter 5 (and in example box 5.2).
  2. dr_stage2_subject**.nii.gz: There is one nifti file per subject, which contains the spatial maps estimated in stage two of the dual-regression. Each file is four dimensional, where the first three dimensions are spatial x, y and z directions, and the fourth dimension includes as many maps as there were in the group-ICA input. These maps are the subject specific-version of the group-ICA components. The values are betas (parameter estimates) from the multiple regression in stage two of the dual regression. These are the maps that should be used for subsequent group-level analyses.
  3. dr_stage2_subject**_Z.nii.gz: These maps contain the same information as the previous maps, but the values in these maps are z-statistics, rather than betas. The difference between these options is that the z-statistics are normalised by the residual within-subject noise.
  4. dr_stage2_ic++.nii.gz: These maps contain the same information as dr_stage2_subject**.nii.gz, but are saved out in a different way. There is one four dimensional file for each ICA component, and each of these files contains one map per subject in the fourth dimension. This way of saving out the stage two dual regression maps is helpful to perform group-level analyses using randomise in FSL.
  5. dr_stage3_ic++_tstat??.nii.gz: the output of "stage 3", i.e. files created by running randomise. Group-level analyses are performed separately for each group-ICA component. You will get one set of statistical output files per group-ICA component, and, within that set of statistical output files, one t-stat (etc.) per contrast in the cross-subject contrast file (design.con). The corresponding corrected (1-p) p-value images are called *corrp*.
  6. dr_stage3_ic++_tfce_corrp_tstat??.nii.gz: These statistical maps are output as 1-p for convenience of display, so for a result to be significant at p<0.05 level, the 1-p value must be over 0.95. If the maximum value in the image is below 0.95, you know that nothing survived thresholding.

** = subject number, in the order of the list entered into the dual regression command.

++ = ICA component number, in the order of the group-ICA maps entered into the dual regression command.

?? = contrast number, in order of the contrasts that were entered into the general linear model used for the group-level analysis.

Visualising group-level dual regression results

A dual regression analysis is used to map group-ICA results back into individual subjects data, e.g. in order to examine between-group difference in ICA networks. We are using data from 12 subjects including six patients with a tumor and six healthy controls.

The directory you have downloaded was created by running the following command (do not run this again):

dual_regression groupICA15/melodic_IC 1 \
  design/unpaired_ttest.mat design/unpaired_ttest.con 5000 \
  groupICA15.dr `cat input_files.txt`

The corrected p-value output images from stage 3 (actually 1-p, for convenience of display) are in files groupICA15.dr/dr_stage3_ic00++_tfce_corrp_tstat??.nii.gz, where the ++ means any one of the 15 group-level components (number 00 to 14) and the ?? relates to the contrast number. To view the results from the dual regression analysis, run:

fsleyes -std groupICA15/melodic_IC \
  -un -cm red-yellow -nc blue-lightblue -dr 4 15 \
  groupICA15.dr/dr_stage3_ic0007_tfce_corrp_tstat3.nii.gz \
  -cm green -dr 0.95 1 &

Make sure you are viewing them over the appropriate volume of melodic_IC (i.e. set the volume of the melodic_IC image to 7, which is the number of the results image we loaded). The dual regression result is very small (because we only had 12 subjects and therefore not much statistical power), to find it please go to voxel location [63 81 54]. You may want to change the minimum threshold at the top to 0.9 to show the results at a slightly more lenient p-value. Note that the results are for contrast 3 (tstat3.nii.gz), which is the comparison of healthy controls minus tumor patients.

Data credits

We are grateful to Natalie Voets for the data that was used in this example.