Is subtracting spmT_0001.nii images valid?

1

Working on my dissertation, and I have used Nipype to analyze my data.
I have (simplified) first-level conditions defined like so:

#                                       A1  A2  A3    A4   B1  B2  B3    B4
cont1 = ['A1 > A4', 'T', cond_names, [ 0.5,  0,  0, -0.5,   0,  0,  0,    0]]
cont2 = ['B1 > B4', 'T', cond_names, [   0, 0,   0,    0, 0.5,  0,  0, -0.5]]

What I would like to do is find the statistical differences between these contrasts. I have an spmT_0001.nii for each contrast after the second-level analysis.

Is it valid to do something like:

fslmaths con_0001/task-ao/spmT_0001.nii -sub con_0002/task-ao/spmT_0001.nii -thr 1.96 alt_diff_pos.nii

to find areas of statistically significant differences?

If not, is there a way to do this without completely reinventing the wheel? Participants have four runs each, and onset times differ for each, so first-level analysis isn’t too cut and dried either (e.g. as in this nilearn example where onsets are the same for everyone.

Is there a way I could dump all the first-level run-0{1,2,3,4}-SPM.mat files for each subject into, nilearn or SPM, define a contrast between my contrasts and get a t-image?

Thanks for any advice!

2

Hi,

  1. Sorry I am a bit confused, can you elaborate more on the structure of your data?:
  • Are A[1-4] and B[1-4] different conditions or different runs?
  • If A* and B* are conditions, are they all present in a single run?
  • Do these all come from the same session?
  • Why are different onsets a problem? Each run should have its own design matrix anyway.
  1. In general, I would work with effect size / beta maps instead of t-maps until you are ready to make inferences.
  2. It sounds like your end contrast would be the same as [ 0.5, 0, 0, -0.5, .5, 0, 0, -0.5].

Best,
Steven

3

Hi @Steven , thanks for the assist.

  1. Each subject has four runs in a single session. Conditions A1-4 occur in a block, and B1-4 occur in block. Eg. A1,A2,A3,A4 are 2sec videos presented in an 8sec period. The contrast A1-A4 shows repetition suppression, as does B1-B4.
    E.g.
ready | prompt | A1 | A2 | A3 | A4 | Fixation |
1s    | 2s     | 2s | 2s | 2s | 2s | 8-12s    |

  1. Okay, I think I better understand, You’re talking about the beta maps for each condition of each run of each subject, and finding statistical differences between the conditions, which then becomes a t-image.

  2. I think you’re right. When I’ve been thinking contrasts, I’ve been thinking ‘only positive’, but the results may be positive and negative, which would give A>B and A<B.
    How would I assess which is greater? Would that be a visual comparison of activations for A>fixation vs. B>fixation?

Thanks for the insight, and helping me think through this!

–
John

4

Ah I see, seeing it as repetition suppression makes more sense now.

Also I think I forgot to multiply the B conditions by -1 in my initial answer, so it would be

[.5 0 0 -.5 -.5 0 0 .5] because it’s the A suppression pair > B suppression pair, but I have to think more about this.