Most likely to Nick S, unless anyone else would like to chime in...
I'm gearing up for some extensive use of FS... it's downloaded, installed, and initial tests run OK... so far with minimum of pain (yay team!).
Now... I need to discover how to feed in data other than raw scanned files, because we want to try several alternatives to the initial processing, which we have in AFNI format, ac-pc aligned with iso-sided voxels but not 1 mm.
I suspect I'll end up scouring the recon-all script, but a few hints would be helpful, mainly aimed at (a) taking advantage of processing that we have done by other methods, (b) minimizing resampling and (c) winding up with surfaces and segmentations that are most readily overlayed with our other processed data, tracings etc.
1. If we want to feed in already ac-pc aligned, brain-extracted input, at what stage of recon-all should we be looking to supply it? (I can figure out where in the directories to put the data, but not sure on the stage of processing to start at).
2. It used to be that FS's surface pipeline didn't insist on Talairaching the voxel data, and skipping that resulted in output that is true to original space. Is that still the case?
3. Does current FS proper require 256x256x256 1mm data, or can it take any shape "brick" and voxel size (ie: without resampling) these days?
4. If FS *does* want to work with 1mm voxels only: It used to be that though FS wanted 1mm input voxels, you could feed in some other sizes by do-it-yourself creation of the COR images, and FS was fine... is that still the case (but now with mgz format)?
4. Is it the case that the subcortical voxel-based parcellation *does* require the Talairaching step for correspondence with the probability atlas?
5. The surface and voxel-based pipelines are separate, not even sharing the white segmentation step, right?
Thanks in advance,
Graham
Graham,
It's tough to say just where you can begin processing with already ac-pc aligned brain extracted non-conformed input, as this newer recon-all processing stream has many interdependencies intended to make it fully automated for 'standard' input. That said, there are many flags allowing custom processing, and the ReconAllDevTable page is the best thing to study (well, actually the recon-all script is best, if you like gory details written in csh). I see you have already created an extension to the ReconAllDevTable. BTW, if you wish to create wiki pages directly on our site, you are welcome to get a wiki account:
http://surfer.nmr.mgh.harvard.edu/fswiki/GettingAccess
Referring to the ReconAllDevTable, OtherUsefulFlags and recon-all pages (and two other references are the scripts/recon-all-status.log and scripts/recon-all.log files for sample subject bert), some special recon-all flags are:
-cm : this forces conforming the rawavg.mgz volume to the smallest voxel size, rather than the default 1x1x1mm
-noaseg : this skips the automatic segmentation step, which would be necessary if not using 1x1x1 conformed images, as the aseg atlas requires a 1x1x1 conformed input
-notalairach : skips creation of transforms/talairach.xfm file, refer to the ReconAllDevTable (or your nice new table) to see impact of not having this.
Note that -notalairach does not skip creation of the file transforms/talairach.lta in the -gcareg (subcortical voxel-based parcellation) step (I'm not sure why this is the case). mri_fill uses talairach.lta unless the -noaseg flag is used (in which case mri_fill just performs a fill using an older algorithm).
So probably the place for you to start, assuming ac-pc aligned brain- extracted data, is to manually run mri_fill to create the filled.mgz volume. Then run the rest of the -autorecon2 steps: -tesselate, - smooth1, -inflate1, -qsphere, -fix, -finalsurfs, -smooth2, -inflate2 and -cortribbon, then -autorecon3.
I don't have experience doing what you are trying to do, so these are suggestions based on my somewhat limited but growing knowledge of the recon-all processing stream. Of course feel free to post any detailed questions, and if you do successfully create a working processing stream, it would be great to create a wiki page add it to our User Contributed Work Flows page:
http://surfer.nmr.mgh.harvard.edu/fswiki/FreeSurferWorkFlows
Nick
On Sun, 2006-05-28 at 13:00 -0700, Graham Wideman wrote:
Most likely to Nick S, unless anyone else would like to chime in...
I'm gearing up for some extensive use of FS... it's downloaded, installed, and initial tests run OK... so far with minimum of pain (yay team!).
Now... I need to discover how to feed in data other than raw scanned files, because we want to try several alternatives to the initial processing, which we have in AFNI format, ac-pc aligned with iso-sided voxels but not 1 mm.
I suspect I'll end up scouring the recon-all script, but a few hints would be helpful, mainly aimed at (a) taking advantage of processing that we have done by other methods, (b) minimizing resampling and (c) winding up with surfaces and segmentations that are most readily overlayed with our other processed data, tracings etc.
- If we want to feed in already ac-pc aligned, brain-extracted input, at
what stage of recon-all should we be looking to supply it? (I can figure out where in the directories to put the data, but not sure on the stage of processing to start at).
- It used to be that FS's surface pipeline didn't insist on Talairaching
the voxel data, and skipping that resulted in output that is true to original space. Is that still the case?
- Does current FS proper require 256x256x256 1mm data, or can it take any
shape "brick" and voxel size (ie: without resampling) these days?
- If FS *does* want to work with 1mm voxels only: It used to be that
though FS wanted 1mm input voxels, you could feed in some other sizes by do-it-yourself creation of the COR images, and FS was fine... is that still the case (but now with mgz format)?
- Is it the case that the subcortical voxel-based parcellation *does*
require the Talairaching step for correspondence with the probability atlas?
- The surface and voxel-based pipelines are separate, not even sharing the
white segmentation step, right?
Thanks in advance,
Graham
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Nick and Doug:
Thanks guys for the thoughtful comments -- I need to do some further digesting.
Nick: Yes I'll get a Wiki username.
Both: Looks to me like I really need to understand what is the coordinate system of each input and output file in the pipeline -- this would greatly clarify what data is immediately compatible with what, or what transforms (and resampling tradeofss) are needed.
And yes, after an initial conception that the subcortical seg and surface processing were completely separate, I did notice that output from the subcortical seg feeds into the preparation of volumes for the surface processes. (So thanks Doug for the further comments on that... again, still digesting!).
Graham
At 6/2/2006 09:51 AM, you wrote:
Graham,
It's tough to say just where you can begin processing with already ac-pc aligned brain extracted non-conformed input, [...]
Graham Wideman wrote:
Most likely to Nick S, unless anyone else would like to chime in...
I'm gearing up for some extensive use of FS... it's downloaded, installed, and initial tests run OK... so far with minimum of pain (yay team!).
Now... I need to discover how to feed in data other than raw scanned files, because we want to try several alternatives to the initial processing, which we have in AFNI format, ac-pc aligned with iso-sided voxels but not 1 mm.
I suspect I'll end up scouring the recon-all script, but a few hints would be helpful, mainly aimed at (a) taking advantage of processing that we have done by other methods, (b) minimizing resampling and (c) winding up with surfaces and segmentations that are most readily overlayed with our other processed data, tracings etc.
- If we want to feed in already ac-pc aligned, brain-extracted input,
at what stage of recon-all should we be looking to supply it? (I can figure out where in the directories to put the data, but not sure on the stage of processing to start at).
mri/orig/001.mgz
- It used to be that FS's surface pipeline didn't insist on
Talairaching the voxel data, and skipping that resulted in output that is true to original space. Is that still the case?
It never actually "talairached" anything. It just computed the transform, and it should not hurt to do it with your data. Otherwise, you could just create an identity matrix (but it is not clear that this would really work).
- Does current FS proper require 256x256x256 1mm data, or can it take
any shape "brick" and voxel size (ie: without resampling) these days?
It does not require this as input but it will convert to it. You can put anything into mri/orig/001.mgz, and it will be conformed by the time it gets to orig.mgz
- If FS *does* want to work with 1mm voxels only: It used to be that
though FS wanted 1mm input voxels, you could feed in some other sizes by do-it-yourself creation of the COR images, and FS was fine... is that still the case (but now with mgz format)?
- Is it the case that the subcortical voxel-based parcellation *does*
require the Talairaching step for correspondence with the probability atlas?
It will do it's own.
- The surface and voxel-based pipelines are separate, not even
sharing the white segmentation step, right?
They are not quite that separate. Subcort structures are used to fill in the ventricles, basal ganglia, etc.
doug
Thanks in advance,
Graham
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu