Hi Henk,
Thank you for your help! I have 28 subjects in total, not that many.
On 10/15/2016 04:36 PM, Henk-Jan Mutsaerts wrote:
> Dear Paul,
>
>
>
> This seems interesting! How many subjects have you got?
> I would look if there are biasfields that are larger or smaller than
> usual, and play with the Biasfield Regularization setting in SPM12's
> segmentation, see what works best. E.g., GE 3T has a bit larger
> biasfield, so I disable regularization for this scanner only, for other
> 3T vendors the standard SPM settings work fine.
I'm not sure what you mean with this. What should I look for exactly?
Each T1 should have a certain biasfield? Should I disable the biasfield
regularization?
>
> For your normalization you might want to use a less blurry template, the
> ICBM512 templates are very crisp, so you can experiment trying different
> smoothing settings. But if the segmentations look good already, then you
> may not have to do this.
Where can I find this template? Is it in the spm directory? I find the
segmentation to be blurry also. It's like I'm losing information after
segmentation. If I understand correctly, I can set the smoothing for the
segmentation? What would you suggest if I have 0.7mm data? 1mm?
>
> What I do before DARTEL, is I resample the images myself into standard
> space (using the inverse transformation that SPM used to segment the T1
> image), and then you can choose any resolution you would like. Indeed in
> your case you could go as low as 0.5 mm isotropic (since is subject is
> scanned in a slightly different orientation) as long as computation
> force allows this (it becomes exponentially slower of course). I myself
> usually use 1.5x1.5x1.5 mm resampling for DARTEL and everything else,
> with a 1x1x1 mm T1 and 3x3x3 mm functional data (ASL).
Do you resample the T1 images or the bias field corrected images or both?
>
> If you use the inverse transformation from the segmentation, and you use
> the deformations tool to resample into standard space, this is a very
> good starting point for DARTEL (since all pGM and pWM images are already
> very similar). DARTEL works better with more variance, with is better
> with a larger sample size, so your final normalization results will also
> depend on this.
>
> I have a manual for my ASL processing toolbox (ExploreASL) in which I
> explain these parts with examples, if you're interested. I've processed
>> 2500 3T 3D T1 images with this procedure, but have never done it with
> 7T data, so hence why I'm interested in the results! Let me know if you
> need any further help,
>
Sure, send it over, I'd love to look into it!
> Best, Henk
>
Thanks again!
Regards,
Glad
> --
>
> Best wishes/hartelijke groet,
>
> **
>
> **
>
> *Henk Mutsaerts, MD PhD*
>
> Sunnybrook Research Institute/Academic Medical Center
>
> Phone: +1 647 5754 824 / +31 6 4390 8284
>
> Skype: hj.mutsaerts
>
--
Paul Glad Mihai, PhD
Independent Research Group "Neural Mechanisms of Human Communication"
Max Planck Institute for Human Cognitive and Brain Sciences
Stephanstraße 1A, 04103 Leipzig, Germany
Phone: +49 (0) 341-9940-2478
E-mail: [log in to unmask]
|