Dear Michael,
VBM8 is using the "New Segment" approach only for initial skull-stripping (older versions also used the New Segment segmentations as starting values). The segmentation in VBM8 differs from the "New Segment" method in the following steps:
- core segmentation is completely different. The Adaptive Maximum A Posteriori (AMAP) method is not using any priors for segmentation
- de-noising of data using spatially-adaptive non-local means (SANLM) filter
- skull-stripping using graph cut
- partial volume estimation (PVE) to allow mixed classes
- DARTEL using affine instead of rigidly registered segmentations
Thus, it is expected to obtain deviating results, although I am somewhat surprised that the differences are so huge.
Best,
Christian
PS: I would recommend to apply an absolute threshold (start with 0.1 and increase up to 0.25) for the statistical analysis to prevent that effects are detected outside gray matter or even outside the brain.
On Fri, 20 Sep 2013 22:50:53 +0100, Michael Regner, MD <[log in to unmask]> wrote:
>Hi SPM experts,
>
>I wanted to expand upon a previous post in which I asked about the differences between "New Segment" and "Estimate and Write" as it pertains to a custom-template DARTEL VBM analysis.
>
>I have a sample population of n=132 roughly evenly distributed between control and experimental subjects. I ran two VBM-DARTEL pipelines, with age as a covariate, both on the SAME population:
>1) "New Segment," followed by "Run DARTEL (create Template)," followed by "Normalize to MNI Space" (with 8mm smoothing)
> - This is based on John Ashburner's 2010 tutorial.
>2) "VBM8: Estimate & Write," followed by "Run DARTEL (create Template)," followed by "VBM8: Write already estimated segmentations," followed by an 8mm smoothing.
> - This is based on the VBM8 manual.
>
>The results, attached, are vastly different. Does anyone have insight into this? Is there an obvious error in my methods? My understanding is that pipeline (2) is based on New Segment, so the results should be very similar, correct? Pipeline (1) takes about 20hrs, while pipeline (2) takes about 40 hrs.
>
>Thank you in advance for your help,
>Michael Regner, MD
>University of Colorado School of Medicine
>
|