 |
 |
Dear FSL experts, I am confronted with some problems related to tbss analysis. I have searched the archives for the answers and learned a lot from them. But I got contradictory information regarding one of my questions. I also need some guides for the other issue. Much appreciate your answers！ （1）i want to test the interaction between disease status and age on FA values. It was the design.mat that confused me. There are two groups, disease1 and disease2. In link http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#FEAT_details-6, it tells me to demean the age across all the subjects before splitting it into two EVs. I also searched the archives and one post said to demean age within group as in https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1005&L=FSL&P=R18717&1=FSL&9=A&I=-3&J=on&X=3A86614E58392038A4&Y=qinxu.china%40gmail.com&d=No+Match%3BMatch%3BMatches&z=4. I don’t know which one is proper in my analysis. When to demean EV within group and when to demean EV across all subjects? (2) Regarding correlation between a behavior score and FA values, I used three different design matrixes without very clear knowledge about them, would you give me some advice? The behavior scores are correlated with disease severity, ie, Patients of disease1 are severer with higher scores than patients of disease2. Option 1: Did not model group EV, so there were 3 EVs in the design: EV1, scores, EV2, age (demeaned across all subjects), EV3, education (also demeaned across all subjects), EV2 and EV3 were added as covariates. The contrast was [1, 0, 0] for positive correlation and [-1,0, 0] for negative correlation. I used –D option in randomize to demean the data. I got no results from both the two contrasts. Option 2: modeled the two disease groups with one EV only and padded with 1s. Contrasts were [0,1,0,0] and [0,-1,0,0], this time didn’t use –D option in randomize. I got a result with contrast [0,1,0,0] but not with [0,-1,0,0], it is strange that FA and the behavior scores are positively correlated which means the FAs in some clusters are increased as disease get severer. Is this result reasonable? Option3: Put the groups in separate EVs, padding 1s for disease1 and 0 for disease2, and vice verse. The contrast became [0, 0,1,0,0] and [0, 0,-1,0,0], the third column here was the behavior scores. Run randomise with –D. I got no significant result again. Which one is reasonable? Thanks in advance for the time and kindheartedness. Best, Qin