Hello experts,
I asked a question about this data a short while ago, but have come to realise the problem is more complicated than I'd thought.
I've got patients and controls. Patients are significantly slower on a reaction time (RT) task. A TBSS analysis has shown they also have significantly different AD values *but* the direction of this change is that patient AD is *higher* than control AD. I understand this is "unconventional", as one would typically expect higher AD values to correspond with better tract health (but also get there's various things that can influence these values to change etc. etc. - for the time being I'm interpreting the increase as a sign of damage, especially given the concurrent cognitive deficit).
So I was wanting to correlate the AD with cognition across both groups, but this poses a problem. Assuming the increased AD is indeed driven by injury, we would expect the controls to exhibit a typical positive correlation between AD and RT, but for the patients to potentially have a negative relationship between those variables.
What I've done so far is the simplest thing I could think of to get the ball rolling and try to understand the data a little better. I've correlated AD with RT separately in the two groups. The model set up I used is the same as described in this other jiscmail thread - https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=fsl;73e149f0.1706 (1000 permutations). This gave me two tstat images for each experimental group.
Looking at the TFCE corrected images in the CONTROLS, there were some small regions of significant pixels (i.e. value above >0.95) in the tstat1 image but not in tstat2. To double check the directionality of this, I extracted the mean AD's of those significant regions from the all_AD_skeletonised dataset and ran these in a straight Pearson correlation with the RT outcomes in SPSS. Sure enough it's significant in a positive direction, as would be expected. The highest value in the tstat2 image on the other hand was 0.126, showing an extremely weak influence of any negative relationship.
In the PATIENTS, neither tstat image was significant however examining the highest value in each I think possibly indicates some change of influence: here the maximum tstat1 value is 0.613, while for tstat2 it is now 0.816. I know this is still a fair drop away from p reaching <0.05, but the shift from 0.126 (controls) to 0.816 (patients), combined with the underlying hypothesis, makes me wonder if there is something going on here but if this very simplistic model approach I've used isn't sensitive enough to detect it.
My question would be if there's any way of setting up a model in FSL that would examine both groups at once, while also being sensitive to there possibly being two different relationships going on? Or indeed any other set-up that's better equipped to handle this sort of bi-directional data?
Many thanks in advance,
Iain
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
|