Back to PLS HelpPosted on 05/25/12 10:47:32
Number of posts: 3
I am trying to use the PLSgui to perform and analysis of structural MRI data, in a longitduinal between groups design. I have 20 participants per group per timepoint, and I cannot, unfortunately, work out how I should set up my datamat structures in order to carry out this analysis. If it's any help (or if mixed models are somehow a problem) I can obviously remove the repeated-measures aspect of the design by looking at difference images. Nevertheless even when considering a dataset with one image per subject I'm not grasping things correctly.
I appreciate that my confusion almost certainly stems from an unfamiliarity with this particular technique, so I apologise for the thoroughly basic nature of my questions
My question is (I hope) fairly simple: How do I set up the datamats to perform this analysis?
There is an intriguing post somewhere earlier in the forum abotu not considering the concept of group "until later' (i.e. not at the datamat stage), which leads me to think that every subject should have their own datamat. However, even based on this premise, I am confused, as I then don't see what goes into an individual datamat- would it be a subject's time1 and time 2 scans? Furthermore, I am slightly baffled as to why WM and GM are considered different conditions in the datamat, but that is a tangent, I think.
I had previously been carrying out SVD of my data using an ungainly concoction of Matlab code, and I am obviously excited at the prospect of being able to use a toolbox to smooth my data-processing pipeline.
Thanks in advance for any advice,
Alexis
Replies:
I'm Online
Posted on 05/25/12 11:31:59
Number of posts: 291
Since it is about brain's anatomical structure, we consider different part of the brain (e.g. WM, GM, etc.) to be the conditions in the datamat.
In order to setup the datamats to perform the Structural analysis, you need to prepare at least 2 types of data. One is the brain region data. This is the binary data include the brain area that you want to view results. The other one is the structural data, which can be mroe than one kind, like WM, GM, etc.). Put al structural data in the data folder, and put the brain region data in the working folder.
Here I give you a simple example:
Under C:\PLSdata, you have:
brainregion.nii
which is the brain region binary data.
Under C:\PLSdata\GroupA, you have the following:
A001_wm.nii
A001_gm.nii
A002_wm.nii
A002_gm.nii
...
A020_wm.nii
A020_gm.nii
Under C:\PLSdata\GroupB, you have the following:
B001_wm.nii
B001_gm.nii
B002_wm.nii
B002_gm.nii
...
B020_wm.nii
B020_gm.nii
Now, launch MATLAB and goto C:\PLSdata, put any string under "Datamat Prefix" (e.g. GroupA), put C:\PLSdata\GroupA under "Dataset Directory".
Then, click "Inupt Conditions", and put the proper condition name and condition filter. e.g. White for name, *_wm.nii for filter. If you have more structural parts, keep add, and then click DONE.
Then, click "Select Subjects", click Add, and you will have subjects A001, A002, etc.
Than, click "Create Datamat". Click Browse and choose brainregion.nii for "Brain Mask File".
Click "Create", and you will get session / datamat for Group A.
Do same thing for Group B.
Posted on 06/06/12 09:32:08
Number of posts: 3
thank you very much for that extremely rapid response.
I tried out what you suggested, and it worked, in so far as the program produced some results.
However, I am slightly concerned about the output, in that only two LVs were generated. I realise that if the 2 groups are taken as wholes then 2 LVs is all I should expect, however, I would have anticipated a decomposition to (a maximum of) 1 LV per datapoint.
I presume that I have misunderstood something about what the PLS output is supposed to represent.
I'd be very grateful for any pointers.
Thanks again,
Alexis
I'm Online
Posted on 06/06/12 09:54:15
Number of posts: 229
Posted on 06/06/12 10:26:27
Number of posts: 3
alexis... if you are getting only 2 LVs, then it would seem you have combined conditions/groups. time1 and time2 should be specified for each group in your datamat, e.g., gm1, gm2 for each group's datamat. This will give you 4 means G1T1,G1T2, G2T1, G2T2, with 3 degrees of freedom, and 3 LVs. nancy
I'm Online
Posted on 07/26/17 10:03:31
Number of posts: 10
alexis... if you are getting only 2 LVs, then it would seem you have combined conditions/groups. time1 and time2 should be specified for each group in your datamat, e.g., gm1, gm2 for each group's datamat. This will give you 4 means G1T1,G1T2, G2T1, G2T2, with 3 degrees of freedom, and 3 LVs. nancy
Dear nlobaugh,
I am trying to use the PLSgui to perform a longitudinal analysis of structural MRI gray matter images for music training with the " Seed Pls " method. I have 27 participants in my control group, and 29 participants in my training group. The training lasted for 6 months. For these 2 groups, each participant was performed 3 times MRI scans (at the beginning of the training, the ends of the 6 months training, and 9 months after the end of the training, respectively). How should I array the datamats to perform this analysis?
According to your proposal, time1, time2 and time3 should be specified as conditions named gm1, gm2 and gm3 for each group. Over all, I set up my datamats as follows:
Under C:\PLSdata\Control, I have the following:
C001_gm1.nii
C001_gm2.nii
C001_gm3.nii
C002_gm1.nii
C002_gm2.nii
C002_gm3.nii
...
C027_gm1.nii
C027_gm2.nii
C027_gm3.nii
Under C:\PLSdata\Train, I have the following:
T001_gm1.nii
T001_gm2.nii
T001_gm3.nii
T002_gm1.nii
T002_gm2.nii
T002_gm3.nii
...
T029_gm1.nii
T029_gm2.nii
T029_gm3.nii
Is it right? .
or Is it like I should make a datamat that include all subjects' data of control or training group in time 1 as group1 and time 2 as group2, and so on? As a result, I can get 6 groups, in which 3 groups correspond to the 3 timepoints data of control, and another 3 groups correspond to the 3 timepoints data of training.
Thanks in advance for any advice.
guo o
Login to reply to this topic.
