Print

Print
Down at the bottom of the user interface of the segmentation is an option called "Clean Up".  You could try setting this to "Thorough Clean" to see what happens.

The problem parts are voxels with the same image intensity as GM that occur in parts of the head where you may expect to find GM.  If these few voxels are the only mistakes (unlikely), it would mean the segmentation is pretty accurate overall.

Best regards,
-John


On 1 July 2016 at 15:03, Sam Berens <[log in to unmask]> wrote:
Dear all,

I am having a problem with the 'New Segment' module (or just 'Segment' in SPM12). In a sample of 40 participants, the routine works without any significant issues in all but one case. In the problematic case, the routine is incorrectly classifying some of the meninges as grey matter (see the attached 3D render of this problematic volume)...

I have tried increasing the MRF parameter (to 2) and performing a 'Thorough Clean Up'. However, neither of these changes have made a difference (apart from inappropriately cutting away genuine grey matter in the former case). I guess I could further increase the strength of the MRF parameter but, as I do not know the details of the model that this parameter is implemented in, I am kind of shooting in the dark.

Any help and suggestions would be greatly appreciated. For your information, listed below are the parameters that I have been using to run the segmentation each time.

Number of Gaussians:
    Grey matter = 2;
    White matter = 2;
    CSF = 2;
    Bone = 3;
    Soft tissues = 4;
    Air = 2;

Sampling distance = 1;
   
<All other parameters maintained at default values>.

Best wishes,
Sam
SamBerens.co.uk


​