Does 3dLMEr -vVar option accept surface (.niml.dset)?

AFNI Message Board
1

Hi there,
I had a problem running 3dLMEr on surface files. My goal is to investigate the effect of baseline surface. So I put covariate BL as vVar, but I don't know if surface vVars are accepted. Alternatively I could convert all surface to volume to do this, but I am just wondering if surface vVar is possible.
Here is my code:

-model 'TREAT*BL+AGE+TREAT*VIS_ORDER+(1+VIS_ORDER|Subj)'   \
         -vVars 'BL' \  # volume vars: BL (each subject has only 1 baseline surface, is surface vVar causing errors?)
         -qVars 'AGE,VIS_ORDER' \ # quantitative variables: age and visit order
         -gltCode PLA.BL        'TREAT : 1*PLA BL : ' \ # effect size of BL in PLA group
         -gltCode DRUG_PLA.BL     'TREAT : 1*PLA -1*DRUG BL : ' \ # compare effect size of BL
         -gltCode DRUG_PLA.VIS_ORDER       'TREAT : 1*PLA -1*DRUG VIS_ORDER : ' \ # compare effect size of visit order
	     -dataTable @delta_cort_thick_bl_lh.txt

Here is the error. I have successfully run the similar test using similar code on volume data so my diagnosis for error culprit is vVars.

~~~~~~~~~~~~~~~~~~~ Model test failed  ~~~~~~~~~~~~~~~~~~~
Possible reasons:

0) Missing R packages lme4, lmerTest, and phia. 
1) Inappropriate model specification with options -model, or -qVars.

2) Incorrect specifications with -gltCode.

3) inconsistencies in the data table.
4) Inconsistent variable names (e.g., typos, case sensitive characters).

** Error: 
   Quitting due to model test failure...
2

To better evaluate the statistical framework, could you clarify the following aspects of the experimental design:

  • Data acquisition: Please provide some background information about the scanning protocol for each participant.
  • Model structure: What is the nature of the variables included in the group-level model (i.e., which are treated as within-individual vs. between-individual factors)?
  • Covariate rationale: What is the specific justification for including variables such as visit order and age in the final model?

Gang Chen

3

Thanks Gang for replying!
My data are surfaces (.niml.dset files) generated from processing MP-RAGE sequence T1 images. All surfaces are std.20 mesh with cortical thickness values at each vertex.
I am trying to model the effect of baseline cortical thickness (z-score against a normative set) on the change of cortical thickness in follow-up visits relative to baseline (∆Thickness). Input variables are:

AGE: numerical, age at baseline visit
TREAT: group-level factor (DRUG and PLA)
BL: Baseline thickness z score, surfaces, each individual has 1
VIS_ORDER: numerical, each individual has precisely 2 visits, so VIS_ORDER have value 1, 2

The only within-individual variable is the visiting order (follow-up 1 and follow-up 2). Between-individual factors are AGE, BL. I want to compare the effect of BL between DRUG and PLA, so I had a TREAT*BL term. VIS_ORDER marks the year after the beginning of the treatment. AGE is the age at baseline visit (visit 0), used as a demographic variable.

Updates since last post: running same 3dLMEr model on surface converted volumes also gave me the same error. However, if I remove VIS_ORDER from random factor: (1+VIS_ORDER|Subj) -> (1|Subj), the 3dLMEr ran successfully ran on volumes, but surfaces still threw an error message:

[1] "Great, test run passed at voxel (1, 1, 1)!"
[1] "Start to compute 1 slices along Z axis. You can monitor the progress"
[1] "and estimate the total run time as shown below."
[1] "03/10/26 10:49:33.788"
Error in rbind(inData, array(0, dim = c(fill, nF))) : 
  number of columns of matrices must match (see arg 2)
Execution halted

Thanks in advance for your guidance!

4

Based on your current setting, the model should be specified as follows:

-model 'TREAT*BL+AGE+TREAT*VIS_ORDER+(1|Subj)'

For a more detailed breakdown of model specification and syntax within 3dLMEr, I recommend checking out this blog post.

Gang Chen

5

Hi Gang,
This model specification indeed ran successfully on volume data (see update in previous reply). However, running it on surface still gave the error message (see below). Is this because of -vVars being set as surface? removing vVar entirely did run through successfully on surface.

[1] "Great, test run passed at voxel (1, 1, 1)!"
[1] "Start to compute 1 slices along Z axis. You can monitor the progress"
[1] "and estimate the total run time as shown below."
[1] "03/10/26 10:49:33.788"
Error in rbind(inData, array(0, dim = c(fill, nF))) : 
  number of columns of matrices must match (see arg 2)
Execution halted
6

I will take a close look at the situation and see if I could fix the issue without requesting for your data and script.

Gang CHen

1 Like
7

Thanks a lot Gang! After some digging in, I found the bug is line#1291 of 3dLMEr.R, where the padding dimension did not consider the concatenated vVar. I added nF <- dim(inData)[2] to update the dimension and the program successfully ran through. Please verify if that is the correct fix whenever it is convenient for you!

8

Thanks for sharing your finding and the fix! That line is indeed problematic for surface data. Does the new result match the volumetric one?

Gang Chen

9

Hi Gang, the results are consistent with the volume outputs.
Some minor problems with the output:
The output 1D file doesn't have the correct subbrick labels or p values or FDR associated with stat bricks.
The output is a .1D file which needs to be converted to dset using ConvertDset in order to visualize in suma. Would be nice if the program outputs a dset file directly readable by suma.