Hi Elisa (and Joost)

It is definitely (*) possible to feed custom segmentations through the FreeSurfer stream and have surfaces be constructed.

I can sketch the general idea here, but will need to check the system we actually used for all the detail.

Basically, given a GM segmentation you need to construct synthetic 'brainmask.mgz', 'filled.mgz', and 'wm.mgz' volumes. For the most part, you can use monochromatic intensity values in each constituent lh and rh segmentation (i.e. for the brainmask.mgz all the rh gray matter voxels can be set to something like 81, all the lh voxels to 82; the white matter rh voxels to 111 and the lh voxels to 112 -- or something thereabouts). Similarly for the 'filled.mgz' and the 'wm.mgz' volumes -- although the intensities in these were a bit different. I'll have to check.

Once you've created these, you essentially start the recon-all stream quite a bit "downstream". I've forgotten where exactly, but can look. Although where we jump started the stream back then probably doesn't translate directly to today's FreeSurfer.

Having said all this, note the following:

1. We implemented this several years ago on what was current then for FreeSurfer. The recon-all stream has evolved quite a bit since then and our specific "hacks" might not be exactly applicable anymore. Especially as regards at what "downstream" point to start FreeSurfer. We haven't used this system since that work finished.

2. The 'pial' surfaces did not reconstruct correctly. Since our interest at the time was limited to the gray/white junction, this was fine. I suspect that using a monochromatic intensity for the wm values was probably the problem. I remember experimenting with intensity gradients applied across the wm volumes, but don't think any results were satisfactory for the 'pial' surfaces.

If you'd like more info, feel free to ping me directly.

Regards -=R

(*) perhaps 'definitely' is too strong. It is mostly possible... :-P

Elisa Dell'Oglio wrote:

Hi Rudolph,

I was wondering if you could give me any hint regarding how to use an apriori GM segmentation in freesurfer. I basically already have a good segmentation of GM vs WM and CSF from spm and would like to use this towards the computing cortical thickness.

Thanks, Elisa

2009/10/7 Bruce Fischl <fischl@nmr.mgh.harvard.edu mailto:fischl@nmr.mgh.harvard.edu>

Hi Elisa,

Rudolph Pienaar (ccd) put together something for this I think.

cheers
Bruce



On Wed, 7 Oct 2009, Elisa Dell'Oglio wrote:

    Hi all,

    I have a question regarding cortical thickness estimation in
    freesurfer. I
    was wondering whether it is possible to input in freesurfer the
    gray matter
    segmentation as an a priori.
    Specifically, I run freesurfer on a set of cases on which I
    previously run
    spm and obtained a good discrimination between GM, WM and CSF.
    I would like to use these segmentation instead to proceed with
    manual edits
    on freesurfer segmentation output for gray matter segmentation, that
    sometimes was not segmented correctly. Would this be possible?

    Thanks a lot,
    Elisa