Mark,
It is possible then that I might be confused. I thought I had previously
read a few posts concerning cluster thresholding which stated that clusters
must have at least 26 voxels. Hmmm, OK.
Could you please explain though, what you mean by "voxels are considered
connected"? What does this connectivity mean in regards to a large cluster,
say 3000 voxels? If this is a complicated topic, is there another link you
could provide?
Thanks,
Carlos
On Tue, 1 Dec 2009 21:20:25 +0000, Mark Jenkinson <[log in to unmask]> wrote:
>Hi,
>
>There is no pre-defined minimum size threshold. It depends on
>the voxel size, the z-threshold and the measured smoothness
>from the data. Are you confusing this with the connectivity which
>is always 26? (that is, voxels are considered connected if they are
>in the 26-voxel neighbourhood of each other - that is, all neighbours
>within a 3x3x3 neighbourhood)
>
>The thresholding is actually done using the specified p-value
>which is a function of the z-threshold, measured smoothness,
>and number of voxels in the cluster. The exact formula is
>quite complicated. If you want, you can find more information at:
> http://imaging.mrc-cbu.cam.ac.uk/imaging/PrinciplesRandomFields
>
>All the best,
> Mark
>
>
>On 1 Dec 2009, at 19:42, Carlos Faraco wrote:
>
>> I have some results results that were cluster thresholded in Feat
>> using the
>> default z and p-values, and of course the minimum cluster size of
>> 26. The
>> results of this analysis are null. I thresholded the z-stat image
>> from the
>> stat directory in fslview using a z of 2.3. There are clusters of
>> contiguous
>> voxels that have much more than 26 voxels. What is the reason why
>> this is
>> happening? Is there anything else that is being done when the cluster
>> program thresholds at the specified minimum cluster size?
>>
>> Thanks,
>>
>> Carlos Faraco
>>
|