hi Donald
> The downside of this approach is that you've still got a voxel
> threshold and extent.
>
> My impression was Steve was wanting something that was independent of
> a specific voxel threshold. For example, you could use .01 and X
> voxels and find clusters that are less than 0.05 OR you could use .001
> and Y voxels to find clusters that are less than 0.05. Both would
> control the cluster level results, but might have different clusters.
> Neither is more correct than the other.
You're exactly right on this point, reflecting that the vast majority
of cluster-correction approaches require some (arbitrary) way of
defining "clusters" (i.e. a cluster-defining threshold). The only
exception I'm aware of is threshold-free cluster enhancement (TFCE)
which is implemented in FSL, and described by Smith & Nichols (2009):
Smith SM, Nichols TE (2009) Threshold-free cluster enhancement:
Addressing problems of smoothing, threshold dependence and
localisation in cluster inference. NeuroImage 44:83-98.
However, it's also that case that when doing a standard
"cluster-defining threshold + cluster correction" approach, SPM
requires a bit more work than other programs—e.g. in FSL, you can
simply specify the voxel and cluster thresholds, and get a
cluster-corrected image.
Jonathan
--
Dr. Jonathan Peelle
Department of Neurology
University of Pennsylvania
3 West Gates
3400 Spruce Street
Philadelphia, PA 19104
USA
http://jonathanpeelle.net/
|