Hi FSL experts,
I have a questions regarding specifying group level design in FSL:
here are snapshots of the design: https://www.dropbox.com/sh/jrx1avhq2suhdkm/AADbvY9doVAt2AW8QFuku6WWa?dl=0
In my project, I asked a group of healthy controls and a group of Major depressive disorder (MDD) patients to partake in an fMRI study (I am assuming the details of the task are not relevant here). I used FSL throughout to carry out my analysis, and I did an initial analysis to examine group-level differences between healthy and MDD groups using an independent-samples t-test. In this analysis, I specified two groups (i.e. different variances were calculated for each group) in FSL, and not just added group as an EV in the group-level design. Age, gender, motion were included as covariates. With this design, I did not see any voxels which survived correction in the between-group contrast.
I then went on to examine how depression severity (rather than depression diagnosis as the previous analysis) may explain variances in the contrast of interest. Specifically, across both healthy and depressed individuals, I wanted to see how scores on the Beck’s Depression Inventory (BDI) may explain task-related activations. I came to this decision mainly because the diagnosis-based division of groups was somewhat diffused – I had several subjects in the healthy control group who had BDI scores in the level of “mild depression” (BDI>8) and a few subjects in the MDD group who did not have high BDI scores (<8). So I wanted to test the effect of BDI scores as it might be a better operationalisation of depression severity.
Nonetheless, diagnosis-based group membership has to be included in the analysis, since there maybe other differences owing to this membership – whether the subject is on medication, therapy, etc. So I specified the attached model in FSL (image attached, full design has 38 rows).
As you see in the design, I have specified two groups, added in Age, Gender and Motion (FD) as well as BDI scores into the model. Attached is also the correlation matrix of the design. I did not mean-center the BDI scores as I am interested in the mean-effect when BDI = 0. There seems to be some correlation between BDI scores and the separate group variables, which is understandable. The contrasts I estimate are also attached.
Specifically, to look at the mean effect for healthy controls I set EV1 as the only EV with a weight of 1 (COPE 3 in the screenshot). I do the same for MDD, and I use the contrast [1 1] to get the average mean effect across both groups (COPE 8). To get the average (negative) effect of BDI across both groups, I use [-1 -1] for EVs 5 and 6 (COPE 7). I also test the interaction effect for BDI x group.
At this point, I would like to understand if the design I have specified is valid in asking the questions I am interested in (mean effect for individual group as well as across group, effect of BDI), or if there is any error I am making in the design specification. I have roughly used the group-level analysis used in this paper to guide my design specification (PDF attached to this email):
Satterthwaite, T. D., Kable, J. W., Vandekar, L., Katchmar, N., Bassett, D. S., Baldassano, C. F., ... & Gur, R. E. (2015). Common and dissociable dysfunction of the reward system in bipolar and unipolar depression. Neuropsychopharmacology, 40(9), 2258.
I will not lengthen this any more than it already is, and I cannot thank you enough for your input.
Best,
Avijit
|