Which SPM updates are you using? It is possible that these problem
regions fall outside the tissue probability maps, in which case a more
recent set of SPM8 updates may help.
Best regards,
-John
On Fri, 2009-10-23 at 14:54 +0100, Jonathan Peelle wrote:
> In general we've been having very good results with the new
> segmentation in SPM8 (e.g., subjects in which segmentation failed
> using the old segmentation now seem to be working ok). However, I've
> run into a few subjects in which tissue clearly outside the brain is
> getting mis-segmented as GM/WM/CSF (see attached).
>
> As a first attempt at fixing this, I reduced the probability of these
> regions being GM/WM/CSF in the tissue probability maps by:
>
> 1) Taking an MNI-space brain mask and dilating it slightly, making
> sure it wouldn't cut off any GM/WM/CSF from TPM.nii
>
> 2) Using this dilated brain mask to set the TPM maps for GM/WM/CSF to
> 0 for areas clearly outside the brain
>
> 3) Re-computing the 'other' TPM (class 6) to be 1 - (sum of the other 5)
>
> This reduced the amount of mis-classified tissue, but did not
> completely solve the problem. I don't know how much of an issue this
> extra tissue is for creating DARTEL templates (I'm guessing not much);
> however, I'm interested in estimates of GM/WM/CSF across subjects, and
> so would like to remove contribution of tissue that is clearly
> non-brain. (The interest in some measure of CSF volume is also one
> reason I don't want to skull-strip before segmenting.)
>
> The only other idea I can think of is to use a dilated brain mask to
> go through each subject and set regions clearly outside of the brain
> to 0, visually inspecting each to ensure the mask was appropriate. I
> wonder if this seems reasonable, and what other, more elegant
> solutions I might be missing....any thoughts or suggestions would be
> much appreciated.
>
> Jonathan
--
John Ashburner <[log in to unmask]>
|