Hello-
I have a question re 3dmerge and 3dresample. I am analyzing some data from a patient population (schizophrenia) and wanted to blur the data before doing parametric tests to increase SNR. Previously I had always blurred the results following parametric testing. My EPI data is 28 slices, 201 images with a voxel size of 4 x 4 x 5. I ran the data through standard pre-processing (despike, tshift, registration) and then used the command "3dmerge -doall -1blur_rms 4" on the data- however, i realized that i do not have isotropic voxels and so the blurring is not equal in all directions. i have tried to resample the data down to 1 x 1 x 1 voxels but i keep getting malloc errors even when using a computer that has 4 g for memory. i was not sure how big of a problem this was so i created a volume with 9 images from 201 and then 1) used 3dresample to go to 1 mm voxels and used 3dFWHM to estimate smoothness (See RSM below) 2) used 3dmerge to do the 4mm blur on the 4 x 4 x 5 data, resampled to 1 mm isometric and then assessed smoothness (GB4.RSM below) and 3) reampled to 1 mm voxels, then blurred with a 4 mm kernel using identical commands as in 2 (RSM.GB4 below). ALl estimates of smoothness were performed only on the area encompassing the brain. As is evident below, the results were very different across the two comparisons- i looked on the message board and could not find any messages addressing this issue- i also looked at the example from one of the how to's (5 i think) and it looks like the shell script does this blurring even though the voxels were square in this experiment as well.
so my question is is this a big deal and wouldn't it be better to blur on isotropic voxels? i imagine that how much difference it will make will depend on the difference in the dimensions of the voxel and how large your kernel is- if it is a big deal, can this be incorporated into the 3dmerge command so that the blurring occurs as if the voxels are isotropic much like the existing dxyz=1 command is for clustering? when i tried this with the blur the images became extremely smooth.
Thanks for taking the time! it is always appreciated!
1) [amayer@amayer TESTING_BLUR]$ more RSM
Gaussian filter widths:
sigmax = 1.38 FWHMx = 3.26
sigmay = 1.59 FWHMy = 3.75
sigmaz = 1.37 FWHMz = 3.24
2) [amayer@amayer TESTING_BLUR]$ more GB4.RSM
Gaussian filter widths:
sigmax = 1.96 FWHMx = 4.63
sigmay = 2.23 FWHMy = 5.24
sigmaz = 1.81 FWHMz = 4.27
3)[amayer@amayer TESTING_BLUR]$ more RSM.GB4
Gaussian filter widths:
sigmax = 4.50 FWHMx = 10.59
sigmay = 4.40 FWHMy = 10.37
sigmaz = 4.28 FWHMz = 10.08