Hi guys
I've read Neuroimage. 2009 Aug 15;47(2):602-10. doi: 10.1016/j.neuroimage.2009.04.053. Epub 2009 Apr 23.
and thought I could do something similar: I've got GM prob maps, Hurst component map (parameterisation of rsfMRI data) and intensity normalised PET images from 3 different groups. I've chosen structural PLS with 3 conditions but then I got stuck at which analysis to use. I went for Behav PLS and entered group as behavioural variable and got results that matched the pathology. But then I couldn't understand how to relate the different conditions(image types) to the LVs that I was looking at. Anybody able to help me with the interpretation or indeed advise me as to whether behav PLS was the correct method.
thanks in advance
Anna
If I've understood correctly, you've entered 1,2,3 for the behaviour variable - this shouldn't have worked, as all subjects in group 1 would have a value of 1, and the correlation could not be calculated..
can you provide some additional details??
nancy
Hi guys
I've read Neuroimage. 2009 Aug 15;47(2):602-10. doi: 10.1016/j.neuroimage.2009.04.053. Epub 2009 Apr 23.
and thought I could do something similar: I've got GM prob maps, Hurst component map (parameterisation of rsfMRI data) and intensity normalised PET images from 3 different groups. I've chosen structural PLS with 3 conditions but then I got stuck at which analysis to use. I went for Behav PLS and entered group as behavioural variable and got results that matched the pathology. But then I couldn't understand how to relate the different conditions(image types) to the LVs that I was looking at. Anybody able to help me with the interpretation or indeed advise me as to whether behav PLS was the correct method.
thanks in advance
Anna
HI Anna - interesting idea. Can you give a bit more detail on how you use "group" as a measure? I would have used two contrasts that code for group membership.
Hi both
Well yes, I probably haven't done it correctly at all :)
So I assumed that I would enter group (1, 2, or 3) for each of the filename entries in one column for condition1, condition2 and condition3 for all 34 subjects (I don't have even numbers in each group unfortunately) so it would be
subj1 condi1 group1
subj2 cond1 group2
subj3 cond1 group3
.
subj1 cond3 group1
subj2 cond3 group2
subj3 cond3 group3 etc
Doing this I got a highly significant LV1 showing me a network that I would expect given the pathology of the 3 groups and a non significant LV2
but I have no idea if that is the correct thing to do! I wanted to find brain LV's that would differentiate the groups in terms of the weightings of the different image types e.g in one network it loaded more on the Hurst maps and PET than on the GM maps in Group2 or it loads equally on both Hurst and GM in Group3.
NB if I choose Mean centered PLS then it does give me a duplicate permutation error message :)
The row order of your behavior data should be in:
subj1 condi1 group1
subj2 cond1 group1
subj3 cond1 group1
.
subj1 cond2 group1
subj2 cond2 group1
subj3 cond2 group1
.
subj1 cond1 group2
subj2 cond1 group2
subj3 cond1 group2 etc
i.e. The row order must be in "subject in condition in group".
Hi both
Well yes, I probably haven't done it correctly at all :)
So I assumed that I would enter group (1, 2, or 3) for each of the filename entries in one column for condition1, condition2 and condition3 for all 34 subjects (I don't have even numbers in each group unfortunately) so it would be
subj1 condi1 group1
subj2 cond1 group2
subj3 cond1 group3
.
subj1 cond3 group1
subj2 cond3 group2
subj3 cond3 group3 etc
Doing this I got a highly significant LV1 showing me a network that I would expect given the pathology of the 3 groups and a non significant LV2
but I have no idea if that is the correct thing to do! I wanted to find brain LV's that would differentiate the groups in terms of the weightings of the different image types e.g in one network it loaded more on the Hurst maps and PET than on the GM maps in Group2 or it loads equally on both Hurst and GM in Group3.
NB if I choose Mean centered PLS then it does give me a duplicate permutation error message :)
hmm - I have a thought!
Treat this as a single group with each image as a 'condition' then do a behavior PLS with the grouping contrasts as the behaviors. You probably want to use orthogonal coding:
v1: 2 -1 -1
v2: 0 1 -1
then what you should get for each LV is the weights for condition (images) will give you the image weights that differentiate the groups
It might work!
Randy
Sounds like a plan. However, its Friday afternoon here and I'm about to go on holiday so you will have to watch this space for a week or so :)
Many thanks to you all
Anna
Hi
I finally got around to running this design. Just to check; I had 2 conditions Grey Matter maps (GM) and Hurst Maps (HM), then I put in 2 columns of behavioural variables in the form you suggested 2 -1 -1 equating to NC - (D1+D2) and 0 1 -1 equating to D1-D2. Where D=disease.There were 34 subjects in total so 68 rows the first 34 being condition 1 and the second 34 being condition 2. 1000 permutations and 100 bootstraps. Output 4 LV pairs
Is it possible to upload a pdf attachment? Anyway my interpretation would be looking at each image LV and each Behav LV:
LV1 (p=0.0001) brain scores separates NC from D1and D2 in condition 1 (GM) r=-0.72
LV2 (p=0.159) brain scores separates D1 from D2 in condition 1 (GM) r=-0.63
LV3 (p=0.352) brain scores separates D1 and D2 in condition 1 (GM) r=0.71 and NC from D1and D2 in condition 2 (HM) r=-0.57
LV4 (p=0.734) brain scores separates D1 and D2 in conditon 2 (HM) r=-0.58.
For the Behavioural LVs (this bit I don't understand so much)
LV1 behav scores separates NC from D1 and D2 in condition 1 (GM) r=-0.89
LV2 behav scores separates D1 from D2 in condition 1 (GM) r=-0.71 and NC from D1 and D2 in condition 2 (HM) r=-0.62
LV3 behav scores separates D1 and D2 in condition 1 (GM) r=0.55 and NC from D1 and D2 in condition 2 (HM) r=-0.73.
LV4 behav scores separates D1 and D2 in conditon 2 (HM) r=-0.94.
Baycrest is an academic health sciences centre fully affiliated with the University of Toronto
Privacy Statement - Disclaimer - © 1989-2024 BAYCREST HEALTH SCIENCE. ALL RIGHTS RESERVED