>Hi. I've been using snpm to perform second-level inference, and have
>recently turned to using the combined voxel-cluster inference options as
>well. I am a bit confused about a result I obtained recently. Using a
>threshold of p=0.05 corrected, and using theta=0.5 (default weighting of
>voxel-cluster combination), and using meta-combining of combining
>functions, I get a number of "significant" clusters. One such cluster
>that is reported, and which survives the filtering (resulting in a t image
>being written that includes these voxels) has the following statistics:
>
>extent: 53 voxels
>cluster-level p-corrected: 0.1091
>combo cluster-level p corr: 0.0646
>combo cluster-level w: 3.89
>voxel level p FWE corr: 0.0410
>
>The Critical W (bottom of results display) is 4.1558, and the p-corr
>meta-combined threshold I specified was 0.05. So to my understanding,
>this particular cluster does not meet the requirements for significance.
>Am I misunderstanding? Is there a reason this cluster is reported in the
>results section and also survives in the filtered statistic image?
In your case, the cluster is reported in your output because its peak
survived the FWE p<0.05 threshold. But in terms of its spatial extent,
there was no significant activation detected at this cluster. Furthermore,
because the spatial extent of the cluster was not large enough, I suspect
the combined method failed to detect the activation. In regular SnPM, if
the cluster extent p-value or the peak intensity voxel p-value is smaller
than 0.05 (or whatever the value you choose), then the cluster is included
in the glass brain. In the combined analysis in SnPM, a cluster is included
in the glass brain if any of cluster, peak-intensity, or combined p-value
is smaller than 0.05.
I hope this answers your question.
-Satoru
Satoru Hayasaka ==============================================
Post-Doctoral Fellow, MR Unit, UCSF / VA Medical Center
Email: shayasak_at_itsa_dot_ucsf_dot_edu Phone:(415) 221-4810 x4237
Homepage: http://www.umich.edu/~hayasaka
==============================================================
|