Hopefully this is an easy question for you. Could you use a func-bucket dataset which contains 64 coefficients (beta coefficients) the same way as a 3d+time dataset in 3dTcorr1D? How do you convert the two where the func-bucket dataset becomes a 3d+time with 64 coefficient timepoints. I want to correlate coefficients instead of timepoints unlike a normal timeseries. I also thought of dumping out the coordinates and 64 coefficients and use Matlab to correlate and use 3dUndump to ‘put’ it back into AFNI. What I would like is to use clusterize program and cluster the correlation coefficient data–would a statistic be necessary for me to do this? Any ideas and comments will help.
A beta weight dataset should behave the same way as an MRI
time series dataset (though you might have to consider a
demean operation, depending on how the correlations are
Is there something that you are trying that is not working?
Would you provide some corresponding commands?
Here is my program where I resample the ROIs into the space of the coefficient file: DeconIMGAMPPI+tlrc. Then I dump out the middle frontal coefficients
into a file called tempmiddle and subsequently transpose the matrix. Finally, I correlate every middle set of coefficients with every set in the Insula giving 1024 Briks but there were I believe 1883 coefficient sets–is 1024 a cutoff? Anyway, it seemed to work giving different correlation values in the Insula ROI which makes me happy. I just need to prove the correlations are correct. I used a pearson correlation. Thanks for your help! -Linda
3dresample -master DeconIMGAMPPI+tlrc -prefix Rmiddlenew -input Rmiddle+tlrc
3dresample -master DeconIMGAMPPI+tlrc -prefix RInsulanew -input RInsula+tlrc
3dmaskdump -mask Rmiddlenew+tlrc -noijk DeconIMGAMPPI+tlrc>tempmiddle.txt
1dtranspose tempmiddle.txt tempmiddle-transpose.txt
3dTcorr1D -pearson -mask RInsulanew+tlrc -prefix middleresults DeconIMGAMPPI+tlrc tempmiddle-transpose.txt
To be sure, what is the output of:
1d_tool.py -show_rows_cols -infile tempmiddle.txt
or maybe more directly:
3dROIstats -quiet -nzvoxels -mask Rmiddlenew+tlrc Rmiddlenew+tlrc
However it seems like what you want could more directly
come from 3dAutoTcorrelate. Consider something like:
3dAutoTcorrelate -polort -1 -mask_source RInsulanew+tlrc
-mask Rmiddlenew+tlrc -prefix test.AC DeconIMGAMPPI+tlrc
What I want is correlation between voxels of two different ROIs. What I get out of my analysis is a BRIK for a given voxel in the Rmiddle correlated to each voxel in the Insula. These BRIKs I would like to put each into an FDR giving a list which can be thresholded by q-value leaving the remaining “active” voxels- ones with significant correlations. One can then convert the BRIK values to binary and then one ‘adds’ a given voxel across BRIKs. This results in a ‘count’ for each voxel in the Insula where the larger the count the more significant that voxel is for the correlation of the two brain regions. What is hoped is that it will result in a cluster in the Insula and simiilarily in the Rmiddle frontal. There should be a connection between these regions. Unfortunately, the Rmiddle frontal is huge with 2883 voxels and when I run this there is a limit of 1025 BRIKs to be stored in the bucket. Can this be changed or else I may have to remove parts of the ROI which is the standard one in Talariach space. Although I find a lot of zero voxels. I don’t know if they can be deleted without the other voxels losing there ‘place’ in the list so to speak. Thanks for all your help. -Linda
I have not yet found where that 1024 limit is coming from.
But in the meanwhile, it does seem like that 3dAutoTcorrelate
command should do the same thing. Run it and just compare
the first few resulting volumes (hopefully the orders match).
3dAutoTcorrelate -polort -1 -mask_source RInsulanew+tlrc \ -mask Rmiddlenew+tlrc -prefix test.AC DeconIMGAMPPI+tlrc
A related question, why there are so many voxels in each ROI?
What is the resolution of the data?