Back to PLS Helpnifti files in PLS
Posted on 02/02/07 12:20:17
Number of posts: 100
Author: Agatha (---.csbmb.Princeton.EDU)
Date: 01-08-07 11:43
hi all ~
i have data that i analyzed in AFNI - have a group effect that i like, but now want to analyze the data using the PLS gui. i have used 3dAFNItoNIFTI to convert the files, then expand_nii to convert to img/hdr files, then ran the analysis. the results are garbage - and now i am trying to troubleshoot the problem. since the data are fine when viewed by AFNI, i expect there is either a problem with the conversion to NIFTI or with the data itself (preprocessing perhaps).
i have three questions...
1. i've tried viewing the converted files using load_nii.m/view_nii.m - and also using SPM (on the .img/.hdr equivalents of the .nii files). these files are quite dim, with lots of variability in value - basically i can't view them properly b/c the intensity scale for graphing doesn't capture the values in the image. there are a couple of outliers in the background that set the range of the colorbar to much larger than the range of the brain intensity values!!! when i view the .nii files in AFNI, they look much clearer. is there any way of adjusting the colorbar, or scale used for the colormap in view_nii.m so that i could better look at these images? is there a different application that might be more appropriate? i am not sure what happens in the conversion... and why the files look fine in AFNI but not in SPM or view_nii.
2. the above problem occurs (i think) b/c my data are normalized to percent signal change - is this something that is a problem for PLS analysis (i.e., do i need more range in values? note that i used a mask in the PLS analysis to select the brain area - this mask was created by converting a mask created in AFNI into NIFTI and then into img/hdr files)?
3. when i view the converted .img/hdr files (via expand_nii.m) in SPM i noticed that the origin is listed as 0 256 -0 - does this mean that the origin is lost? i thought that the NIFTI format would preserve this information! will my PLS analysis have lost the origin as well? (making interpretation of coordinates senseless)
thanks for any help or advice ...
agatha
Date: 01-08-07 11:43
hi all ~
i have data that i analyzed in AFNI - have a group effect that i like, but now want to analyze the data using the PLS gui. i have used 3dAFNItoNIFTI to convert the files, then expand_nii to convert to img/hdr files, then ran the analysis. the results are garbage - and now i am trying to troubleshoot the problem. since the data are fine when viewed by AFNI, i expect there is either a problem with the conversion to NIFTI or with the data itself (preprocessing perhaps).
i have three questions...
1. i've tried viewing the converted files using load_nii.m/view_nii.m - and also using SPM (on the .img/.hdr equivalents of the .nii files). these files are quite dim, with lots of variability in value - basically i can't view them properly b/c the intensity scale for graphing doesn't capture the values in the image. there are a couple of outliers in the background that set the range of the colorbar to much larger than the range of the brain intensity values!!! when i view the .nii files in AFNI, they look much clearer. is there any way of adjusting the colorbar, or scale used for the colormap in view_nii.m so that i could better look at these images? is there a different application that might be more appropriate? i am not sure what happens in the conversion... and why the files look fine in AFNI but not in SPM or view_nii.
2. the above problem occurs (i think) b/c my data are normalized to percent signal change - is this something that is a problem for PLS analysis (i.e., do i need more range in values? note that i used a mask in the PLS analysis to select the brain area - this mask was created by converting a mask created in AFNI into NIFTI and then into img/hdr files)?
3. when i view the converted .img/hdr files (via expand_nii.m) in SPM i noticed that the origin is listed as 0 256 -0 - does this mean that the origin is lost? i thought that the NIFTI format would preserve this information! will my PLS analysis have lost the origin as well? (making interpretation of coordinates senseless)
thanks for any help or advice ...
agatha
Replies:
Re: nifti files in PLS
Posted on 02/02/07 12:20:39
Number of posts: 100
Author: Jimmy (---.rotman-baycrest.on.ca)
Date: 01-08-07 12:40
There may be several reasons for your unexpected results. With my limited knowledge, I try to answer some of your questions:
> 1. i've tried viewing the converted files using
> load_nii.m/view_nii.m - and also using SPM (on the .img/.hdr
> equivalents of the .nii files). these files are quite dim, with
> lots of variability in value - basically i can't view them
> properly b/c the intensity scale for graphing doesn't capture
> the values in the image. there are a couple of outliers in the
> background that set the range of the colorbar to much larger
> than the range of the brain intensity values!!! when i view the
> .nii files in AFNI, they look much clearer.
As the author of MATLAB NIfTI Tools, what I can tell you is that the "load_nii.m" truly load data in .nii file. I have not go deep into AFNI or SPM program, so I have no knowledge why AFNI can bring you a much clearer look.
> is there any way of
> adjusting the colorbar, or scale used for the colormap in
> view_nii.m so that i could better look at these images? is
> there a different application that might be more appropriate? i
> am not sure what happens in the conversion... and why the files
> look fine in AFNI but not in SPM or view_nii.
In terms of the adjustment of the appearance, "view_nii" provided limited numbers of colormap selection, contrast scrollbar, and brightness scrollbar. It may give you a better look, but really depends the data itself.
> 2. the above problem occurs (i think) b/c my data are
> normalized to percent signal change - is this something that is
> a problem for PLS analysis (i.e., do i need more range in
> values? note that i used a mask in the PLS analysis to select
> the brain area - this mask was created by converting a mask
> created in AFNI into NIFTI and then into img/hdr files)?
I can not answer this question.
> 3. when i view the converted .img/hdr files (via expand_nii.m)
> in SPM i noticed that the origin is listed as 0 256 -0 -
"expand_nii" does nothing else but save each time frame into individual .nii file (i.e. 4D->3Ds).
> does
> this mean that the origin is lost? i thought that the NIFTI
> format would preserve this information! will my PLS analysis
> have lost the origin as well? (making interpretation of
> coordinates senseless)
NIFTI format should contain origin information, and "load_nii" truly extracted origin information from .nii file.
Date: 01-08-07 12:40
There may be several reasons for your unexpected results. With my limited knowledge, I try to answer some of your questions:
> 1. i've tried viewing the converted files using
> load_nii.m/view_nii.m - and also using SPM (on the .img/.hdr
> equivalents of the .nii files). these files are quite dim, with
> lots of variability in value - basically i can't view them
> properly b/c the intensity scale for graphing doesn't capture
> the values in the image. there are a couple of outliers in the
> background that set the range of the colorbar to much larger
> than the range of the brain intensity values!!! when i view the
> .nii files in AFNI, they look much clearer.
As the author of MATLAB NIfTI Tools, what I can tell you is that the "load_nii.m" truly load data in .nii file. I have not go deep into AFNI or SPM program, so I have no knowledge why AFNI can bring you a much clearer look.
> is there any way of
> adjusting the colorbar, or scale used for the colormap in
> view_nii.m so that i could better look at these images? is
> there a different application that might be more appropriate? i
> am not sure what happens in the conversion... and why the files
> look fine in AFNI but not in SPM or view_nii.
In terms of the adjustment of the appearance, "view_nii" provided limited numbers of colormap selection, contrast scrollbar, and brightness scrollbar. It may give you a better look, but really depends the data itself.
> 2. the above problem occurs (i think) b/c my data are
> normalized to percent signal change - is this something that is
> a problem for PLS analysis (i.e., do i need more range in
> values? note that i used a mask in the PLS analysis to select
> the brain area - this mask was created by converting a mask
> created in AFNI into NIFTI and then into img/hdr files)?
I can not answer this question.
> 3. when i view the converted .img/hdr files (via expand_nii.m)
> in SPM i noticed that the origin is listed as 0 256 -0 -
"expand_nii" does nothing else but save each time frame into individual .nii file (i.e. 4D->3Ds).
> does
> this mean that the origin is lost? i thought that the NIFTI
> format would preserve this information! will my PLS analysis
> have lost the origin as well? (making interpretation of
> coordinates senseless)
NIFTI format should contain origin information, and "load_nii" truly extracted origin information from .nii file.
Re: nifti files in PLS
Posted on 02/02/07 12:20:58
Number of posts: 100
Author: Agatha (---.csbmb.Princeton.EDU)
Date: 01-08-07 12:58
Thanks Jimmy.
Is the PLSgui able to read in nifti files directly (during datamat creation) or do the files always have to be ANALYZE format (converted via expand_nii.m)?
Date: 01-08-07 12:58
Thanks Jimmy.
Is the PLSgui able to read in nifti files directly (during datamat creation) or do the files always have to be ANALYZE format (converted via expand_nii.m)?
Re: nifti files in PLS
Posted on 02/02/07 12:21:20
Number of posts: 100
Author: Jimmy (---.rotman-baycrest.on.ca)
Date: 01-08-07 13:29
> Is the PLSgui able to read in nifti files directly (during
> datamat creation) or do the files always have to be ANALYZE
> format (converted via expand_nii.m)?
For E.R.fMRI & Blocked fMRI, you can read in NIfTI files directly, both 3D & 4D.
For PET, however, you must extract 4D NIfTI to 3D NIfTI. By the way, the file format generated using "expand_nii" is NIfTI, rather than ANALYZE. This reason is explained in PLSgui User's Guide at:
http://www.rotman-baycrest.on.ca/pls/UserGuide.htm#_Toc146699241
Date: 01-08-07 13:29
> Is the PLSgui able to read in nifti files directly (during
> datamat creation) or do the files always have to be ANALYZE
> format (converted via expand_nii.m)?
For E.R.fMRI & Blocked fMRI, you can read in NIfTI files directly, both 3D & 4D.
For PET, however, you must extract 4D NIfTI to 3D NIfTI. By the way, the file format generated using "expand_nii" is NIfTI, rather than ANALYZE. This reason is explained in PLSgui User's Guide at:
http://www.rotman-baycrest.on.ca/pls/UserGuide.htm#_Toc146699241
Re: nifti files in PLS
Posted on 02/02/07 12:21:40
Number of posts: 100
Author: Randy (---.dsl.bell.ca)
Date: 01-08-07 21:36
Jimmy wrote:
>
> > 2. the above problem occurs (i think) b/c my data are
> > normalized to percent signal change - is this something that
> is
> > a problem for PLS analysis (i.e., do i need more range in
> > values? note that i used a mask in the PLS analysis to select
> > the brain area - this mask was created by converting a mask
> > created in AFNI into NIFTI and then into img/hdr files)?
>
>
> I can not answer this question.
If I understand what you are doing, Agatha, the AFNI normalization is done somewhat differently than for PLS. If I recall correctly, AFNI adjusts signal across the entire run whereas PLS does local adjustments only. If you can save the data from AFNI w/o the normalization, this would be prefereable. Otherwise you could have %signal change data normalized to %signal change from reference scan onset.
Date: 01-08-07 21:36
Jimmy wrote:
>
> > 2. the above problem occurs (i think) b/c my data are
> > normalized to percent signal change - is this something that
> is
> > a problem for PLS analysis (i.e., do i need more range in
> > values? note that i used a mask in the PLS analysis to select
> > the brain area - this mask was created by converting a mask
> > created in AFNI into NIFTI and then into img/hdr files)?
>
>
> I can not answer this question.
If I understand what you are doing, Agatha, the AFNI normalization is done somewhat differently than for PLS. If I recall correctly, AFNI adjusts signal across the entire run whereas PLS does local adjustments only. If you can save the data from AFNI w/o the normalization, this would be prefereable. Otherwise you could have %signal change data normalized to %signal change from reference scan onset.
Re: nifti files in PLS
Posted on 02/02/07 12:21:59
Number of posts: 100
Author: Agatha (---.csbmb.Princeton.EDU)
Date: 01-09-07 13:52
randy ~ yes that's correct - in AFNI i've normalized (within each run) - and hence running the PLS renormalizes to the reference scan (but this shouldn't matter). i'll try it using raw signal rather than %signal change - although i had thought it would not matter since % signal change is just a linear transformation of the signal.
on that note are there an other steps that are disruptive to PLS analysis ..my preprocessing pipeline includes the usual more or less so... but perhaps i've missed something (these data are rather fragile both for motion and a fast-paced design)
motion correction
smoothing (6 mm)
talairach normalization
normalize (% signal change)
detrend
thanks for the advice
agatha
Date: 01-09-07 13:52
randy ~ yes that's correct - in AFNI i've normalized (within each run) - and hence running the PLS renormalizes to the reference scan (but this shouldn't matter). i'll try it using raw signal rather than %signal change - although i had thought it would not matter since % signal change is just a linear transformation of the signal.
on that note are there an other steps that are disruptive to PLS analysis ..my preprocessing pipeline includes the usual more or less so... but perhaps i've missed something (these data are rather fragile both for motion and a fast-paced design)
motion correction
smoothing (6 mm)
talairach normalization
normalize (% signal change)
detrend
thanks for the advice
agatha
Re: nifti files in PLS
Posted on 02/02/07 12:22:18
Number of posts: 100
Author: Randy (---.rotman-baycrest.on.ca)
Date: 01-09-07 14:51
hi again - I am not sure if the %signal change will affect your results, but do let us know what happens. As for the rest:
> motion correction
mandatory
> smoothing (6 mm)
if this is ~twice your voxel size, okay. Anything more and you will loose sensitivity
> talairach normalization
okay so long as the normalization works well. PLS is very good at detecting registration artefact.
> normalize (% signal change)
not necessary
> detrend
also not necessary unless you event window is very long. If it is, then test for the trend and remove it *only* if its there.
Date: 01-09-07 14:51
hi again - I am not sure if the %signal change will affect your results, but do let us know what happens. As for the rest:
> motion correction
mandatory
> smoothing (6 mm)
if this is ~twice your voxel size, okay. Anything more and you will loose sensitivity
> talairach normalization
okay so long as the normalization works well. PLS is very good at detecting registration artefact.
> normalize (% signal change)
not necessary
> detrend
also not necessary unless you event window is very long. If it is, then test for the trend and remove it *only* if its there.
Re: nifti files in PLS
Posted on 02/02/07 12:22:55
Number of posts: 100
Author: Agatha (---.csbmb.Princeton.EDU)
Date: 01-09-07 15:16
got it. just one last question...
if the AFNI talairaching is unreliable - is there some template that i can rely on? (e.g., SPM?) i will need to do some kind of normalizatino to get all subjects into the same space!
agatha
Date: 01-09-07 15:16
got it. just one last question...
if the AFNI talairaching is unreliable - is there some template that i can rely on? (e.g., SPM?) i will need to do some kind of normalizatino to get all subjects into the same space!
agatha
Re: nifti files in PLS
Posted on 02/02/07 12:23:13
Number of posts: 100
Author: Randy (---.rotman-baycrest.on.ca)
Date: 01-09-07 17:32
AFNI does do an automated Talairaching, which seems to be pretty good. We have used SPM for this, with no nonlinear warping, or FSL's affine transform (FLIRT). I am not sure if there is a big difference between these three approaches. We tend to stick to afffine transforms only to minimize signal distortion from interpolation. There is a potential for added registration artifact between subjects, but thats a reasonable trade-off, I think.
cheers
Randy
Date: 01-09-07 17:32
AFNI does do an automated Talairaching, which seems to be pretty good. We have used SPM for this, with no nonlinear warping, or FSL's affine transform (FLIRT). I am not sure if there is a big difference between these three approaches. We tend to stick to afffine transforms only to minimize signal distortion from interpolation. There is a potential for added registration artifact between subjects, but thats a reasonable trade-off, I think.
cheers
Randy
Login to reply to this topic.
