Hi Ged, Thomas, everyone,
First of all, thanks Ged for making my toolbox compatible with SPM5,
I'll post an updated version on my website soon.
Thomas, I'd like to better understand the pros and cons of the two
solutions (masked contrast images versus variance smoothed t-maps).
Am I right to assume that variance smoothing is done *before* SPM
calculates corrected p-values? How can variance smoothing increase
statistical power? One might assume that smoothing reduces the
variance in one region and increases the variance in another, so both
an increase or a decrease of statistical power is conceivable. If
there is a tendency towards higher power, this sounds to me as if the
generally lower variance in white matter helps to artificially
decrease the variance in gray matter. I might be wrong as I do not
exactly know how variance smoothing is implemented, nor do I know
about the details of the nonparametric statistic. Is there a specific
reason why you would prefer variance-smoothed t-maps over masked
contrast images?
Thanks,
Matthias
Am 16.03.2006 um 19:29 schrieb Ged Ridgway:
> Hi, Matthias, John, everyone,
>
> I think Matthias's approach is very interesting, and a very brief
> look at my own data suggests that significant clusters within the
> brain do shift around a little, but not as much as I feared they
> might.
>
> I was wondering how non-parametric methods (e.g. SNPM) relate to
> this issue -- would the same problem or any similar problems be
> expected?
>
> Many thanks,
>
> Ged.
>
>
> Matthias Reimold wrote:
>> Hi Ged,
>>> The trouble is, if this same effect is shifting peaks around
>>> within the brain too -- and it presumably will, if, for any
>>> reason, certain areas are lower variance than others, which seems
>>> quite plausible, especially for anatomical variability -- then
>>> the effect won't be noticed, and can't be "fixed" by masking.
>>> This seems very worrying.
>> I do not understand why t-blobs should be located outside the
>> brain, but obviously they are. As to t-blobs shifting around
>> inside the brain and a suggestion how to fix them, you may be
>> interested in my paper / toolbox:
>> http://homepages.uni-tuebingen.de/matthias.reimold/mci/
>> I'm curious to here if this algorithm was helpful for your
>> problem, too.
>> Regards,
>> Matthias
|