There are 8 Gaussians, each modeled by a mixing proportion, mean and
covariance (one each for GM, WM and CSF, plus four to model everything
else). Because mixing proportions sum to one, then there are
effectively seven degrees of freedom here. So for univariate
segmentation, there are 23 parameters for the MOG.
For multispectral classification, the there will be a few more
parameters used (k-1 + k*n + k*(n^2/2+n/2)), where n is the number of
channels and k is the number of Gaussians (8).
There are also a bunch of parameters to model the bias. The actual
number will depend on the field of view of your image, and some
frequency cutoff that is specified. Note that these parameters are
regularized, so the effective degrees of freedom will be less than the
number of parameters.
The number of voxels used to compute the parameters will vary. To save
time, a bounding box that encloses the brain is determined and only some
of the voxels within this box used for the computations.
All the details are readable in spm_segment.m
Best regards,
-John
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
On Behalf Of Nosarti, Chiara
Sent: Tuesday, February 20, 2007 4:21 PM
To: [log in to unmask]
Subject: [SPM] feature variables in mixture model
Hello
Could anyone tell me (or refer to appropriate reference) how many voxels
are used as the feature variables in the mixture model in SPM2? And how
many parameters are estimated (12 ?)?
Best
Chiara
__________________________________
Dr Chiara Nosarti PhD, Lecturer in Cognitive Neuropsychology, Division
of Psychological Medicine, Room M5.01.05 Section of General Psychiatry,
PO Box 63, Institute of Psychiatry
16 De Crespigny Park, Denmark Hill,
London SE5 8AF, UK
Tel: 0207 848 0133
Fax: 0207 701 9044
|