Dear YAN Chao-Gan,
I saw you posted a script ( y_Smoothest) which calculates the inherent smoothness of a data set. Thanks for publishing that script, but I have a couple of question related to that:
I wonder if one should rather use the residuals of the GLM, i.e. the file ResMS.img or if it is better to use the con image (e.g. con_0004.img) or even the spmT (e.g. spmT_0004.img) images to determine the inherent smoothness? What would you propose?
I calculated the values with your program and then read that SPM itself does calculate FWHM during the results process as well. The values are given on the results sheet of SPM in the Graphics window on the bottom. However these values and the ones I got with your method (irrespective of the image I used, see above) differ in a way. Do you know what you do differently than SPM? I think SPM takes the residuals as described by Kiebel et al., 1999, NeuroImage (http://www.fil.ion.ucl.ac.uk/spm/doc/papers/sjk_robust.pdf), but even if I take the residuals (ResMS.img), I get different values. SPM unfortunately doesn’t show the values for individual masks, but only for the whole brain.
Moreover, I wonder what you would take as input for FWHM as input for alphasim, the mean of FWHMx, FWHMy, FWHMz?
Thanks a lot for your help in advance and sorry for double posting!
Klas
PhD Student
University of Leipzig
Department for Psychosomatic Medicine
Leipzig
Germany
Re
My initial intention to write this script is want to perform the Gaussian Random Field Theory Correction like easythresh in FSL. Easythresh estimated the smoothness from the Z statistical image, so did this script.
However, the y_Smoothest (soon as rest_Smoothest in the new REST) script also could estimate the smoothness from residual file, and this way maybe better than the estimation based on Z statistical image.
According to Steve Smith, “Hi - you're right that easythresh isn't exactly the same, because it uses the zstat image and not the residuals to estimate smoothness.
However in my experience it isn't ever very different.” (https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0902&L=fsl&P=R48770&1=fsl&9=A&I=-3&J=on&d=No+Match%3BMatch%3BMatches&z=4), estimation from Z statistical image is acceptable.
Thus, you can put the spmT image into y_GRF_Threhold, and then it will convert the T image into Z image by calling y_TFRtoZ, and later estimate the smoothness from the Z image. This procedure will be put the new REST which will be released by the end of this month.
This script is based on FSL’s tech report [Flitney, D.E., & Jenkinson, M. 2000. Cluster Analysis Revisited. Tech. rept. Oxford Centre for Functional Magnetic Resonance Imaging of the Brain, Department of Clinical Neurology, Oxford University, Oxford, UK. TR00DF1.] and their program. However, discrepancy of smoothness estimation from AFNI or SPM to FSL may exist, as a recent post in FSL: https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1208&L=fsl&P=R43270&1=fsl&9=A&J=on&d=No+Match%3BMatch%3BMatches&z=4.
Actually, REST AlphaSim accept smooth kernel like [FWHMx, FWHMy, FWHMz], thus you don’t need to do average. And seems AFNI’s 3dClusterSim could take the output from 3dFWHMx directly.
Best,
Chao-Gan