" " wrote:
> Hi ,spmer:
>
> Siddharth Srivastava([log in to unmask]) asked john a question(Fri, 3 Aug 2001 13:04:54). The content as follow:
> <<<
> dear John, spmers,
> I had the following questions regarding the statistical
> tests for a particular experiment i am doing now. My data consists of
> certain texture features derived from MRI and segmentation maps, which are
> combined in a single number at each voxel. I wanted to compare these
> responses across a group of normal controls, and get a spm of where these
> values significantly differ in the patient.
> Currently i am using "1 scan/subject Ancova", with one
> group containing the normals, the the second group containing a single
> patient. The volumes are normalized and smoothed at 8mmFWHM.
> My question to the group is, is this a correct analysis
> for the comparison?
Yes, this looks fine.
>
> Further, i have observed the following. The way the
> texture features are generated yields a sparse volume, and there are cases
> where the values are zero/near zero at that voxel across all normals, and
> substantially large in the patient. But these seem to get excluded from
> the analysis. How does spm deal with those voxels where the response
> variables are zero everywhere in the normals, but large in the patients?
> And, how can i ensure that all the voxels within the brain get analysed?
The p-values generated by SPM in the above analysis reflect the
difference between the patient texture values and the mean 'normal' texture
values - this difference is then divided by the variance of the normal group so
that *significant* differences can be assessed.
If the normal group
values are all zero, however, then the variance of the normal group will
be zero. This is a big problem.
I can only suggest that you re-define your texture features so that they are never
exactly zero - as long as there is some even very small variation in the
normal group you should be OK.
>
> I hope i have been able to put the question clearly.
> thanking in advance,
> Siddharth.
> >>>
> john's response were:
> <<Without seeing your design matrix or data, I couldn't say if "1 scan/subject Ancova" was correct or not.In terms of defining the region of the image that gets analysed, you need to use explicit masking, whereby you provide an image defining the region you want to search. Zero or NaN values identify voxels that are not included in the analysis.>>
>
> >From above message ,I know Siddharth wanted to make analysis by comparing a single patient with a control group (several persons).. I wonder wether SPM can make analysis by comparing a single patient with a control group (several persons)? If can ,how ?
>
> Many thanks.
> Hamte.
> 01-8-13
>
> 保持联系有新招,ChinaRen校友录1000万名同学等着你!
> http://alumni.chinaren.com
> 圆梦北京,搜狐助你
> http://store.sohu.com/sport.htm
--
William D. Penny
Wellcome Department of Cognitive Neurology
University College London
12 Queen Square
London WC1N 3BG
Tel: 020 7833 7478
FAX: 020 7813 1420
Email: [log in to unmask]
URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
|