I have had considerable frustration in trying to break up an activation map with regions of activation that are too extensive. For reasons that are lengthy to explain, I am stuck with the chore of reporting out cluster data (mean signal intensity/Z-scores/xyz CM coordinates) on a map that contrasts brain activity against baseline, at a fairly lenient statistical threshhold. When I use 3dclust, I end up with 10 large clusters that are actually ~30 distinct brain regions, connected in various ways. I have threshholded the data in various ways, but even with relatively aggressive choices for rmm and vmul in my 3dmerge command, and a stipulation for -1erode of 50 (followed by -1dilate), the blobs are still stuck together. If you've have similar problems, can you tell me what you've done to solve the problem?
I have uploaded a sample image to the AFNI server to illustrate the problem, called Encode.MeanJit_Z.thr003_3.8+tlrc. This image is a Z-score map that has been threshholded to a p of .003, a connectivity radius of 3.8 and a min. cluster volume of ~300 mul. The original voxel size is 3.75x3.75x4.2, resliced for group analysis to 1x1x1. The actual command for the volume threshholding is below:
3dmerge -1noneg -1clust 1.1 295 -1erode 50 -1dilate -datum float \
-prefix Encode.MeanJit_Z.thr003_rmm3.8 \
Encode.MeanJit_Z.thr003+tlrc
Thanks in advance,
AndreaB