I have used 3dROIstats to compute the number of non-zero voxels contained within a region-of-interest mask. I am wondering if there is a way to further compute the number of non-zero voxels within this mask that exceed a given size parameter (e.g., 1 mL contiguous). In other words, I would like to compute the number of non-zero voxels only if there are a total of approx. 296 adjoining voxels (at 1.5 isotropic resolution this would equate approx. 1mL contiguous).

Best,
Jesse

Hi, Jesse-

I think 3dClusterize is what you would want to use, maybe prefaced with 3dcalc for simpler usage.

With 3dcalc, make a mask dset of where voxels are nonzero. Then, use 3dClusterize to find the locations in that mask where clusters of voxels are bigger than a certain size; you have to specify how you want your neighborhoods formed (the “nearest neighbor” number, either 1, 2 or 3-- please check the file’s help description if that isn’t familiar), which brick number you want to threshold (the “ithr” is the volume index for thresholding, say the th volume), how you want thresholding done (since your input dset is a mask in this case, then “-1sided RIGHT_TAIL 0.5” should be fine, since you are only interested in values >0.5), and your minimum clustersize in terms of volume (mL) or voxels (number):

``````
# make a mask of the initial dset where values are nonzero
3dcalc \
-a DSET \
-expr 'notzero(a)' \
-prefix DSET_NONZERO

# Make a new dset of only clusters larger than a min size;  I just picked NN=1 here at random, but choose whatever you feel is appropriate
3dClusterize                  \
-inset DSET_NONZERO        \
-ithr 0                    \
-NN 1                      \
-1sided RIGHT_TAIL 0.5               \
-clust_nvox 296            \
-pref_map DSET_NONZERO_CLUSTMAP

``````

and actually, 3dClusterize’s table will tell you how many voxels in total survived (= the sum of all voxels). Otherwise, you can use 3dROIstats for that (perhaps run on a binarized version of the DSET_NONZERO_CLUSTMAP).

–pt

pt this is great insight. Thanks so much for your help!