[VoxBo] cognitive conjunction analysis in VoxBo

[marina bedny]

Thanks Dan, this is really helpful.
& I think this is a really great tool.

m
On Jun 22, 2006, at 12:30 PM, Daniel Y Kimberg wrote:

> Marina Bedny wrote:
>>> have n=2, you can use vbperminfo to do the test in a permutation
>>> framework.  Use the -pm option and specify multiple perm cube
>>> directories.  vbperminfo will create a distribution of "peak minimum
>>> values," a confusing phrase that means for each permutation it finds
>>> the minimum value, across all the effects, in each voxel.
>>
>> Okay, that part is done. The output is in /jet2/mbedny/PolyHom/
>> conj_inc_cons_cont.txt
>> Where do I go from here? The file contains a cut of values, is this
>> the maximum min value across volumes described below?
>
> Yes, I think I'm settled on calling it the "peak minimum."
>
>>> Then it
>>> picks out the maximum min value across the volume.  Using that
>>> distribution, we can pick a threshold to ensure that X% of the time
>>> under the null hypothesis, there will be no voxel that exceeds
>>> threshold for all effects.
>>
>> How do I create the conjunction map you're talking about here? Can I
>> simply set the threshold for each contrast as to what is specified in
>> the output file of what's above and find the overlap? How do I use
>> the vbstatmap & vbmaskmunge to get a final conjunction map?
>
> Yes -- you just need to find voxels that are above threshold for both
> contrasts.  So create each stat map using vbstatmap, and create the
> combined mask by thresholding and quantizing in vbmaskmunge, something
> like:
>
>   vbmaskmunge -t <thresh> -q mymap1.cub -o mask1.cub
>   vbmaskmunge -t <thresh> -q mymap2.cub -o mask2.cub
>   vbmaskmunge -i mask1.cub mask2.cub -o finalmask.cub
>
> Eyeball the masks in vbview to make sure they look like they're
> supposed to, but I think that's the right syntax.
>
>>> Using that
>>> distribution, we can pick a threshold to ensure that X% of the time
>>> under the null hypothesis, there will be no voxel that exceeds
>>> threshold for all effects.
>>
>> Exceeds threshold for all effects or one or more effects? I thought
>> it was the latter?
>
> It needs to exceed threshold for all effects.  The logic of the test
> is that you're finding voxels where it's unlikely for all N maps to be
> above your threshold if there's no effect in any of them.  Using the
> minimum across maps means that you're finding voxels where seeing the
> effect in all maps is unlikely.
>
> Since you only have two maps, that's the best you can do.  If you
> found a threshold at which only one had to be above threshold, it
> would be the peak maximum (instead of peak minimum).  That wouldn't be
> a useful test, since you could do better just by doing the two maps
> independently, and doing a true conjunction test.
>
> dan