3dFWHMx change - perhaps important to YOU

3dFWHMx no longer shows the “classic” (Forman 1995 -http://onlinelibrary.wiley.com/doi/10.1002/mrm.1910330508/full) estimate of the Gaussian FWHM of the noise smoothness. This change means that anyone using the Gaussian-shape model for the spatial correlation of the FMRI noise will have trouble doing so in AFNI – the purpose of this change is exactly to cause such trouble.

I have two reasons for making this change:[ol]
[li] The Forman method estimates the FWHM using statistics of the nearest neighbor first order differences of the noise volume. This method is not reliable unless the noise spatial autocorrelation function (ACF) is exactly Gaussian.
[/li][li] And the ACF of FMRI data is not Gaussian. Therefore, the default method for estimating the noise smoothness is now the ‘-acf’ option, which gives the 3 (a,b,c) parameters of the mixed model for the ACF, plus the FWHM of the model fit.
The old output from 3dFWHMx -acf looked something like this:

 5.3901  5.41111  5.6564     5.48455
 0.0603753  6.5946  8.09281    11.598

In the above, the last number on the first line is the Forman FWHM estimate (5.5 mm). The last number on the second line is the mixed model ACF FWHM estimate (11.6 mm).

The new output (stdout) on the same data looks like so:

0  0  0    0
0.0603753  6.5946  8.09281    11.598

The first line of all 0s is present to provide compatibility with various scripts that expect 2 lines of 4 numbers each to appear on stdout. For example, afni_proc.py expects the 3dFWHMx output to be in this format, and puts these values into the output file blur_est.SUBJECT.1D (where SUBJECT is the subject ID you supplied).

If you really wish to see the Forman estimate of FWHM, then the output of 3dFWHMx -help will tell you how to reveal those numbers. However, we here at AFNI Galactic HQ strongly recommend against using these Forman estimates for any statistical purpose in the analysis of FMRI datasets. The -acf method generalizes the Gaussian model for the ACF, and is strongly to be preferred. If the noise spatial correlation is actually Gaussian-shaped, then the mixed model estimate will reflect that.

At the present time, you can still use the -fwhm option in 3dClustSim to generate cluster-size thresholds for the pure Gaussian ACF model. We do not recommend this, and prefer that you use the -acf option in 3dClustSim. Even better, use the new -Clustsim or -ETAC option in 3dttest++ for generating cluster-size thresholds.