Hi Doug and Freesurfers,
I am using fcseedcor to compute the correlation between two time courses for each subject in my group. The command that I run is as follows:
fcseedcor -s $subject -fsd resting -seed seed1.dat -seed seed2.dat -xreg global.waveform.dat 1 -xreg vcsf.dat 5 -xreg wm.dat 5 -xreg mcprextreg 6 -hpf .01 -lpf .08 -nskip 4 -o cor_s1s2.dat
My first question is about the size of the vcsf.dat and wm.dat files. They seem to be (N+1)xN matrices, where N is the number of timepoints in the signal. Does this mean there are N+1 components for each time point (i.e., N+1 potential regressors to add)? What do they exactly correspond to and how do they relate to the average signal from the ventricles, CSF, and white matter?
My other question is about the number of components to include for vcsf.dat and wm.dat. I have seen 5 recommended in examples, however, five components seem to explain very different amount of variance in different subjects and if I change this number for a given subject, I see substantial changes in the resulting correlation value. I would appreciate any suggestion on how to select an unbiased value for the number of components to include. Could including up to the component at which fixed percentage of cumulative variance is explained be a solution?
Many thanks for your help!
Best, Hamdi
Hi Hamdi, sorry for the delay. Answers below.
On 6/26/18 2:50 PM, Eryilmaz, H. Hamdi wrote:
Hi Doug and Freesurfers,
I am using fcseedcor to compute the correlation between two time courses for each subject in my group. The command that I run is as follows:
fcseedcor -s $subject -fsd resting -seed seed1.dat -seed seed2.dat -xreg global.waveform.dat 1 -xreg vcsf.dat 5 -xreg wm.dat 5 -xreg mcprextreg 6 -hpf .01 -lpf .08 -nskip 4 -o cor_s1s2.dat
My first question is about the size of the vcsf.dat and wm.dat files. They seem to be (N+1)xN matrices, where N is the number of timepoints in the signal. Does this mean there are N+1 components for each time point (i.e., N+1 potential regressors to add)? What do they exactly correspond to and how do they relate to the average signal from the ventricles, CSF, and white matter?
Each ROI consists of a matrix of size Ntp-by-Nvoxels. A PCA is computed from this matrix. The the number of temporal components of the PCA will be either Ntp or Nvoxels, which ever is less. In this case, Nvoxels>Ntp, so you see the Ntp components (each Ntp long). The components are whatever the PCA finds. The mean is removed, so it no component represents the mean.
My other question is about the number of components to include for vcsf.dat and wm.dat. I have seen 5 recommended in examples, however, five components seem to explain very different amount of variance in different subjects and if I change this number for a given subject, I see substantial changes in the resulting correlation value. I would appreciate any suggestion on how to select an unbiased value for the number of components to include. Could including up to the component at which fixed percentage of cumulative variance is explained be a solution?
When I was doing experiments with this years ago, i did not find much of a difference with different numbers of components, but I did the experiments on whole brain resting state, so the ROI-wise correlation may be different. It seemed like 5 components explained the bulk of the variance within the ROI (WM or CSF). You could vary the number of components for each subject. Sorry, I don't have any better guidance on how to go about selecting the number. doug
Many thanks for your help!
Best, Hamdi
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Thanks Doug!
One last question: If I wanted to regress out the average signal from the ventricles and the white matter (instead of using the PCA output), could I directly incorporate that into fcseedcor?
Best,
Hamdi
________________________________ From: freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu on behalf of Douglas Greve dgreve@mgh.harvard.edu Sent: Friday, July 20, 2018 9:23:58 AM To: freesurfer@nmr.mgh.harvard.edu Subject: Re: [Freesurfer] fcseedcor -- pca output for vcsf.dat and wm.dat
Hi Hamdi, sorry for the delay. Answers below.
On 6/26/18 2:50 PM, Eryilmaz, H. Hamdi wrote: Hi Doug and Freesurfers,
I am using fcseedcor to compute the correlation between two time courses for each subject in my group. The command that I run is as follows:
fcseedcor -s $subject -fsd resting -seed seed1.dat -seed seed2.dat -xreg global.waveform.dat 1 -xreg vcsf.dat 5 -xreg wm.dat 5 -xreg mcprextreg 6 -hpf .01 -lpf .08 -nskip 4 -o cor_s1s2.dat
My first question is about the size of the vcsf.dat and wm.dat files. They seem to be (N+1)xN matrices, where N is the number of timepoints in the signal. Does this mean there are N+1 components for each time point (i.e., N+1 potential regressors to add)? What do they exactly correspond to and how do they relate to the average signal from the ventricles, CSF, and white matter? Each ROI consists of a matrix of size Ntp-by-Nvoxels. A PCA is computed from this matrix. The the number of temporal components of the PCA will be either Ntp or Nvoxels, which ever is less. In this case, Nvoxels>Ntp, so you see the Ntp components (each Ntp long). The components are whatever the PCA finds. The mean is removed, so it no component represents the mean.
My other question is about the number of components to include for vcsf.dat and wm.dat. I have seen 5 recommended in examples, however, five components seem to explain very different amount of variance in different subjects and if I change this number for a given subject, I see substantial changes in the resulting correlation value. I would appreciate any suggestion on how to select an unbiased value for the number of components to include. Could including up to the component at which fixed percentage of cumulative variance is explained be a solution? When I was doing experiments with this years ago, i did not find much of a difference with different numbers of components, but I did the experiments on whole brain resting state, so the ROI-wise correlation may be different. It seemed like 5 components explained the bulk of the variance within the ROI (WM or CSF). You could vary the number of components for each subject. Sorry, I don't have any better guidance on how to go about selecting the number. doug
Many thanks for your help!
Best, Hamdi
_______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edumailto:Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
you would do it in fcseed-config by using -mean instead of -pca
On 7/20/18 4:06 PM, Eryilmaz, H. Hamdi wrote:
Thanks Doug!
One last question: If I wanted to regress out the average signal from the ventricles and the white matter (instead of using the PCA output), could I directly incorporate that into fcseedcor?
Best,
Hamdi
*From:* freesurfer-bounces@nmr.mgh.harvard.edu freesurfer-bounces@nmr.mgh.harvard.edu on behalf of Douglas Greve dgreve@mgh.harvard.edu *Sent:* Friday, July 20, 2018 9:23:58 AM *To:* freesurfer@nmr.mgh.harvard.edu *Subject:* Re: [Freesurfer] fcseedcor -- pca output for vcsf.dat and wm.dat Hi Hamdi, sorry for the delay. Answers below.
On 6/26/18 2:50 PM, Eryilmaz, H. Hamdi wrote:
Hi Doug and Freesurfers,
I am using fcseedcor to compute the correlation between two time courses for each subject in my group. The command that I run is as follows:
fcseedcor -s $subject -fsd resting -seed seed1.dat -seed seed2.dat -xreg global.waveform.dat 1 -xreg vcsf.dat 5 -xreg wm.dat 5 -xreg mcprextreg 6 -hpf .01 -lpf .08 -nskip 4 -o cor_s1s2.dat
My first question is about the size of the vcsf.dat and wm.dat files. They seem to be (N+1)xN matrices, where N is the number of timepoints in the signal. Does this mean there are N+1 components for each time point (i.e., N+1 potential regressors to add)? What do they exactly correspond to and how do they relate to the average signal from the ventricles, CSF, and white matter?
Each ROI consists of a matrix of size Ntp-by-Nvoxels. A PCA is computed from this matrix. The the number of temporal components of the PCA will be either Ntp or Nvoxels, which ever is less. In this case, Nvoxels>Ntp, so you see the Ntp components (each Ntp long). The components are whatever the PCA finds. The mean is removed, so it no component represents the mean.
My other question is about the number of components to include for vcsf.dat and wm.dat. I have seen 5 recommended in examples, however, five components seem to explain very different amount of variance in different subjects and if I change this number for a given subject, I see substantial changes in the resulting correlation value. I would appreciate any suggestion on how to select an unbiased value for the number of components to include. Could including up to the component at which fixed percentage of cumulative variance is explained be a solution?
When I was doing experiments with this years ago, i did not find much of a difference with different numbers of components, but I did the experiments on whole brain resting state, so the ROI-wise correlation may be different. It seemed like 5 components explained the bulk of the variance within the ROI (WM or CSF). You could vary the number of components for each subject. Sorry, I don't have any better guidance on how to go about selecting the number. doug
Many thanks for your help!
Best, Hamdi
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu mailto:Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu
-
Douglas Greve -
Eryilmaz, H. Hamdi