Using FSL GLM with ALFF images

, ,
1

I am working on an analysis looking at the association between ALFF and depression change, including head motion as a covariate of no interest. So far, I have used fmriprep to preprocess then XCP-D to generate ALFF images, resampled to MNI152 space, applied a mask, calculated the Z-stat image, and merged into a 4D image so that each volume is one subject’s ALFF Z-stat image.

Then, I am using FSL’s GLM, not within FSL FEAT but the separate tool, to perform Higher Level / non-timeseries design. The # of inputs is # of participants, and then including 3 EVs: column of 1s, column of demeaned % depression change values, column of demeaned head motion values. Then for the contrasts, the contrast of interest is
gm dep change head motion
0 1 0

I save that, which saves the .con and .mat files I need for randomise. Then, I run randomise using the basic structure of randomise -i 4D_merged_image_ALFF_zstat -o zstat_alff_randomise -d <design.mat> -t <design.con> -m mask -n 5000 -x --uncorrp

which outputs a t-stat image, an uncorrected p value image, and a corrected p value image. I am then using that t-stat image as the basis of a TDP exploratory analysis.

  1. Does the above sound like an appropriate use and set up of FSL GLM’s tool?
  2. Is there a way for me to have randomise consider negative t-stat values when creating the p-value image? Right now, it’s a one-sided test, but negative t-stat values are relevant.

Thanks in advance for help!

2

Sorry to tag but you’ve really helped me before on fsl related matters and I am quite lost here as to whether using the FSL GLM outside of FEAT is appropriate here. @paulmccarthy @jsein Thanks in advance if you have the time to help!

3

I can’t comment as to whether the test you are performing makes any sense, but all the GLM tool does is generate a design matrix (and associated contrast / f-test information). So if your analysis warrants the use of randomise, then this sounds fine.

Regarding negative t-statistics, all tests performed with randomise are one-tailed, so if you are interested in the opposite tail, you can just add an additional inverted contrast. If you are interested in both tails, you can apply an f-test to both contrasts.