FSL: repeated measures in FEAT/FSL

Software Support
, ,
1

Summary of the experiment:
We have 3 conditions and 2 measurements per condition (pre and post), in a crossover design with 21 subjects. I was able to perform individual-run analyses and pre vs. post contrasts for each condition. However, I am encountering difficulties setting up a group-level analysis with these repeated measures (there are 6 total observations per subject).

Has anyone faced this issue and managed to plan an acceptable configuration using the FEAT GUI/FSL that could help me?

2

I use nipype to set up the contrast, because of the bunch of columns. I prefer the old fsl wiki:
https://fsl.fmrib.ox.ac.uk/fsl/oldwiki/FEAT(2f)UserGuide.html#Group_Statistics

Tripled Two-Group Difference (“Tripled” T-Test) or Multi-Session & Multi-Subject (Repeated Measures - Three Level Analysis).

1 Like
3

For posterity, here is the link in our new documentation: https://fsl.fmrib.ox.ac.uk/fsl/docs/statistics/glm.html#single-group-three-measurements-tripled-t-test

1 Like
4

Hi! This can be tricky. Do you have a rough matrix or a description of what your outcome contrast would ideally look like? Then I might be able to point you the right direction.

1 Like
5

Thanks! I’m still new to FLAME and higher-level modeling in FSL, so I don’t yet have a clear idea of what the design matrix should look like for this case.

but my current approach…

At the group level, I am using:

  • FEAT → Higher-level analysis
  • Paired t-test (Two-Paired Group Difference)
  • Inputs:
    • COPEs from the Pre condition
    • COPEs from the Post condition
  • Model: fixed effects (since I’m focusing on within-sample inference)

My main contrast of interest is:

  • Post > Pre (and the reverse)

My questions

  1. Is this paired higher-level FEAT setup the recommended way to compare Pre vs Post for a visual task in FSL?
  2. Would there be any advantage to using a different approach (e.g., mixed-effects with FLAME, or modeling the difference already at first level)?
  3. For visual tasks specifically, are there best practices regarding:
  • ROI-based vs whole-brain analysis?
  • Handling multiple runs or visits?

Any feedback, suggestions, or references would be greatly appreciated.
Thanks in advance!