Hi Jiansong,
It isn't generally a good idea to surf different methods until you find
one in which some results are found. As for AlphaSim you may want to
know that:
- It uses Monte Carlo simulations, which is not the same as
non-parametric methods as it may sound at first. The assumptions of
parametric methods are all there, and so, there may not be a clear
advantage over the Random Rield Theory used in SPM. And with RFT, you
can have t and F statistics too, not just Gaussian.
- It computes p-values for the clusters individually on the basis of
their sizes, but there is no FWE correction at the cluster level. This
requires the distribution of the maximum cluster size under the null,
which SPM computes parametrically.
- If you are doing VBM, there is spatial structure (non-stationarity) on
the statistical map, which you probably would like to take into account
(see Hayasaka et al (2004) on Neuroimage and the related toolbox
available for SPM). In AlphaSim, on the other hand, the map is assumed
to be stationary.
- AlphaSim doesn't estimate smoothness of the map, which SPM does using
the residuals of the GLM. This is not quite the same as the external
smoothing applied during the preprocessing, and the difference has a
substantial impact on the inference at the end.
Having said this, it is fine to use AlphaSim if none of these issues are
relevant to your project. Otherwise, I'd probably stick with SPM. I
can't remember the difference between the correction in SPM2 and SPM5
(if I ever knew). In any case, if you have the choice of using different
SPM versions, I'd suggest to stick with the most recent throughout the
whole analysis: SPM8.
Hope this helps!
All the best,
Anderson
On 09/01/12 18:37, Jiansong Xu wrote:
> Dear All:
>
> I have two questions.
>
> 1. Are there any difference in algorithms used for corrections for multiple comparisons between SPM5 and SPM2? My impression is that SPM5 is more restrict than SPM2. For example, if I used height threshold p< .01, a cluster of 100 voxels usually can survive correction in SPM2, but SPM5 needs about 250 voxels.
>
> 2. Is it reasonable to use AlphaSim from AFNI to decide cluster size? I noticed that AlphaSim needs much smaller clusters to survive correction relative to SPM5.
>
> Best Regards
>
> Jiansong
>
> The information contained in this message may be privileged and confidential. If you are NOT the intended recipient, please notify the sender immediately with a copy to [log in to unmask] and destroy this message. Please be aware that email communication can be intercepted in transmission or misdirected. Your use of email to communicate protected health information to us indicates that you acknowledge and accept the possible risks associated with such communication. Please consider communicating any sensitive information by telephone, fax or mail. If you do not wish to have your information sent by email, please contact the sender immediately.
> ===============
> Jiansong Xu, M.D., Ph.D.
> Assistant Professor
> Dept. of Psychiatry
> Yale Medical School
> 1 Church St., Room 729
> New Haven, CT 06519
> Tel: 203-785-5306
> Fax: 203-737-3591
>
>
|