Dear Jodi,
Are you looking at zstat1 from the stats directory?
If so, that is the raw, unthresholded zstat and you shouldn't look at that.
Instead, look at the thresh_zstat1 image in the top-level of the feat directory.
This cluster list looks like what you'd generate yourself from the
command line, not what Feat would generate as there are no p-values
displayed for the clusters. Hence it is not actually giving you just the
significant clusters, instead it is giving you *all* clusters. If you open
the report.html from your Feat analysis and then click on an activation image
you will see the cluster table generated by Feat for that contrast. This uses
information about the smoothness of the residuals in order to calculate the
cluster corrected p-values and therefore eliminate small clusters. You can also
find this information in the cluster_zstat1.txt file in the feat directory.
I hope that this makes things clearer.
If you have just run Feat normally then you should not need to run "cluster"
yourself and you should only see larger clusters (based on the significance
which takes into account the smoothness).
If you don't see this, or you have done something else rather than run
Feat in a straightforward way, then let us know what exactly you are doing
and we'll try to help.
All the best,
Mark
On 12 Jul 2011, at 15:32, Jodi Gilman wrote:
> Hi Mark,
>
> I am using "Thresholding: Cluster, Z threshold = 2.3, Cluster P threshold = 0.05." I am still seeing single voxels what I look at zstat1. Should I be looking at a different output?
>
> When I run the cluster command, I get something that looks like this:
> Cluster Index Voxels Z-MAX Z-MAX X (vox) Z-MAX Y (vox) Z-MAX Z (vox) Z-COG X (vox) Z-COG Y (vox) Z-COG Z (vox)
> 69 64 4.3 54 12 32 52.1 13.1 29.6
> 68 17 3.35 23 46 66 23.4 46 66.4
> 67 15 3.17 30 44 65 30.2 43.4 65.8
> 66 14 3.57 57 92 47 56.9 92.8 46.9
> 65 12 3.46 72 50 27 71.7 50.2 27.2
> 64 11 3.02 12 51 41 12.3 50 40.5
> 63 9 3.82 66 55 32 66.2 55.2 32
> 62 9 3.05 24 47 30 24.9 46 30.4
> 61 8 3.19 60 15 40 59.6 14.5 40.6
> 60 8 3.08 67 70 19 67.5 70.6 19.3
>
> I'm not sure how to actually look at the data with only clusters of >10 voxels.
>
> Thanks,
> Jodi
>
>
> Dear Jodi,
>
> If you are using cluster-corrected statistics then you shouldn't
> get activations from single voxels.
>
> By the way - did your cluster command work in the end?
>
> All the best,
> Mark
>
|