Comparison between the two/three(multiple) groups

[fhaos123]

Hello,
I tried to perform comparisons between multiple groups regarding the CT measure, and the script is as follows:

model_RFT_random=RFT_vertex_analysis(model = dat_beh[,c("Group","sex","age")], 
                               contrast = dat_beh[,"Group"], random = dat_beh[,"sex","age"],surf_data = SPRENG_CTv_smoothed, p = 0.05)

Meanwhile, I have a confusion: when the model selects three variables and the contrast selects Group, even if no random factors are input, this RFT_vertex_analysis automatically selects other variables as control variables (random factors) except Group. Is this explanation true?

Best regards,

[fhaos123]

And how to save brain map at 600 DPI

[CogBrainHealthLab]

I don't think you are using the random argument correctly. This random argument is for specifying random effects (in the context of a linear mixed effects model)— typically in the form of subject IDs.

In the case of

model_RFT_random=RFT_vertex_analysis(model = dat_beh[,c("Group","sex","age")], 
                               contrast = dat_beh[,"Group"], surf_data = SPRENG_CTv_smoothed, p = 0.05)

sex and age would basically serve as control/nuisance variables in this analysis.

Currently, there isn't an option to modify the resolution of the output images generated using plot_surf()

[fhaos123]

Dear @CogBrainHealthLab

Thank you for your reply! I completely understand the usage after your explanation.

The functions FSLRvextract and SURFvextract require specific pipelines (such as FreeSurfer, fMRIPrep, or the HCP pipeline) in order to extract certain metrics. In other words, they depend on a well-defined directory structure.

If additional metrics (e.g., ODI and NDI from the derivatives of the NODDI model, as well as ALFF, ReHo, and FC) are manually resampled from volume space to fs_LR 32k and fsaverage5 using FreeSurfer and Workbench, but the data do not follow the expected directory structure, for example, rootdirectory/ contains multiple subject folders, and each subject folder includes metric files for the left and right hemispheres, then the extraction functions may not work properly. Do I understand it correctly？Do these data can be read using additional tools (cifti, ciftitools and freesurferformats in R) ？Is the recommended process to transform the data into an N × vertices matrix, where N is the number of subjects?

Thanks a lot

[CogBrainHealthLab]

FSLRvextract should work quite flexibly with many different directory structures, since you will be searching, recursively in a given path, for files with matching filenames.

I believe SURFvextract only works on FreeSurfer type of directory structures, since it calls upon FreeSurfer functions.

Running FSLRvextract or SURFvextract is all that is required to obtain a ready-to-analyze data matrix object of N × vertices. You should not need to perform any additional formatting/processing (beyond smoothing) on the data matrix object.

[fhaos123]

Hi,
Conducting the contrast between the two groups, I encountered inconsistent results. Specifically, when I converted the group variable to a factor, the analysis yielded more significant differences. In contrast, using the group variable as a character type produced completely inconsistent outcomes. What is the more appropriate type for categorical variables (e.g., group, sex)? From my understanding, the RFT_vertex_analysis should automatically handle the conversion of such arguments, whether they are in factor or character format.

# factor
demo.DM$group <- factor(demo.DM$group,
                       levels = c("NDPN", "DPN"))
DPN_NDPN_DM_CTv_smoothed = smooth_surf(DPN_NDPN_DM_CTv, 10)
CT_model1=RFT_vertex_analysis(model = demo.DM[,c("group","gender","age","age2")],
                            contrast = demo.DM[,"group"], surf_data = DPN_NDPN_DM_CTv_smoothed, p = 0.05)
CT_model1$cluster_level_results

> CT_model1$cluster_level_results
$`Positive contrast`
 clusid nverts      P     X   Y    Z tstat    region
1      1   9353 <0.001 -33.9 8.2 10.6 16.39 lh-insula
2      2   9361 <0.001  35.7 2.7 12.0 17.76 rh-insula

$`Negative contrast`
[1] "No significant clusters"

# character
demo.DM$group <- as.character(demo.DM$group)
DPN_NDPN_DM_CTv_smoothed = smooth_surf(DPN_NDPN_DM_CTv, 10)
CT_model1=RFT_vertex_analysis(model = demo.DM[,c("group","gender","age","age2")],
                              contrast = demo.DM[,"group"], surf_data = DPN_NDPN_DM_CTv_smoothed, p = 0.05)
CT_model1$cluster_level_results

> CT_model1$cluster_level_results
$`Positive contrast`
[1] "No significant clusters"

$`Negative contrast`
  clusid nverts     P    X     Y     Z tstat              region
1      1    128 0.013 44.9 -28.6 -23.5 -3.32 rh-inferiortemporal

Many thanks!

[CogBrainHealthLab]

Currently the analysis functions are only designed to work with numeric or character (binary) variables and not factor. So the second set of analysis/results should be accurate. The first analysis's results are likely wrong— two huge clusters with all the vertices in the brain

Thanks for pointing this out, we'll add support for factor variables in the next update

[fhaos123]

Thanks for your reply!
I may have one more question that will take a moment of your time. Specifically, how to determine the direction of the contrast (i.e., positive versus negative)? Is the group of the first subject treated as the default reference group in the demo data (corresponding to the first row in surf_data )?

[CogBrainHealthLab]

When you run the analyses on a binarized character variable as a contrast, you will see the following line
The binary variable 'sex' will be recoded with F=0 and M=1 for the analysis

Whichever level that is coded as 1 will show greater cortical thickness in the clusters listed in the $Positive contrast, and lower cortical thickness in the clusters listed in the $Negative contrast.

When in doubt, you can always extract the significant clusters as ROIs and compare them across the levels of the group variable