PLS experts,
Our team has collected some data in the fall and we are analyzing the functional data with PLS. The data was collected from young adults on a Siemes Prisma on a 32-channel coil. We had adults resting or watching a video with 554 volumes of data with TR/TE = 650ms/34.8ms. We used afni_proc.py to preprocess the data:
Hi Gig - have you run univariate analyses in AFNI? If so, do they look okay? I was going to check your results today but my matlab license expired so I will look at it on Monday
Randy,
Thanks for your reply and considering looking at the data over the weekend. We ran a univariate analysis on AFNI today using LTTG as a seed. As expected, there is strong correlation between the bilateral regions. The results can be found here:
https://www.dropbox.com/s/d0439l9eoww7mw3/Screen%20Shot%202017-02-13%20at%205.27.23%20PM.png?dl=0
Our plan was to run mean-centered PLS in order to do multiple voxel extraction for seed PLS. Any insight would be greatly appreciated.
Thank you!
Gigi
Randy,
Thanks for your reply and considering looking at the data over the weekend. We ran a univariate analysis on AFNI today using LTTG as a seed. As expected, there is strong correlation between the bilateral regions. The results can be found here:
https://www.dropbox.com/s/d0439l9eoww7mw3/Screen%20Shot%202017-02-13%20at%205.27.23%20PM.png?dl=0
Our plan was to run mean-centered PLS in order to do multiple voxel extraction for seed PLS. Any insight would be greatly appreciated.
Thank you!
Gigi
Hi again Gigi - I took a look at your results file and there is definitely something weird going on. You have some extreme values in the singular vectors that make me think something is up with how the data are being read in. Can you create a mean datamat and take a look at it? You can save the datamat when you run an analysis.
I tried a few times to create and save the mean datamat but failed. I have been using the most updated PLS on matlab. I have tried using the GUI and script. I included all the sessiondata.mat files for the 5 participants, along with the script for the mean centered analysis and the result.mat in the following Dropbox folder
https://www.dropbox.com/sh/9ob0950htpcbah2/AAC6AmQ10AUwskswNf1PggtIa?dl=0
The pattern of weird results are similar to the results we got from the entire sample of 31. One interesting observation is that the brain scores at run 1 block 9 seems to be where the spikes are. The brain scores are much higher than other blocks when the other blocks are hovering around zero.
I would appreciate your advice on this. Thank you so much for taking a look.
Gigi
Gigi..
If you scale the boostrap ratio plot or the LV plot to 0(Pos.Thresh,Neg.Thresh), you will see that the pattern is essentially noise -
I don't see where/how you are telling AFNI to export the result of all of the regressions, but I suspect that is the source of your problems. The values in the images after all of those regression steps should "make sense" - i.e., you should still see some anatomy. If you don't, then AFNI is the potential source of the error.
Nancy
Thanks Nancy. We are also running mc PLS on non-regressed data to see if the results will end up looking different. However, if the regressions are problematic, we should not expect to see meaningful patterns using 3dGroupInCorr in AFNI, which we did (see earlier post with results from LTTG as seed). So, I am not entirely sure that it is the preprocessing issues.
We'll keep you posted on the results from the non-regressed data.
Gigi
it's more the export to nii that I'm worried about.. AFNI will have no problem working with its own datatype, but if your nii files look like noise in another viewer, PLS won't be able to work with it.. also, sometimes when factors are regressed out, the range is too small -- if you can upload the datamats, we will be better able to see if that is a problem..
Also.. since you are using a binary mask, some of the automatic checking for "bad" data are turned off - you could have one bad dataset that is driving this - do you see similar noise patterns if you allow PLS to threshold the data? If there is a bad dataset, you could get a very sparse datamat.
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