Some important notes on smoothness estimation

Using the 4D residual for smoothness estimation

The R-package (from version 0.1.1) is able to compute the smoothness parameters of the underlying Gaussian Random Field model based on the 4D residual file, which is more accurate than Z-stat image based smoothness. Using the 4D residual for smoothness estimation is the recommended way to assess image smoothness.

The 4D residual file can be obtained:

• FSL: <feat_analysis_folder>/<cope.X.feat>/stats/res4d.nii.gz
• AFNI 3dLME: Use the -resid PREFIX to output the 4d residual file.
• SPM: See this link. (But consider using the pTFCE SPM Toolbox, see below)

Alternatively, you can rely on smoothness parameters estimated by 3rd party software tools and pass the number of voxels to the V, and the number of resels (or resel count) to the Rd parameter of the ptfce function.

These smoothness parameters can be obtained as follows:

• FSL FEAT: use the smoothness information written out by feat into the text file <feat_analysis_folder>/<cope.X.feat>/stats/smoothness.

Please note that the RESELS field of the output is not the number of resels, but the resel size (in voxels). Therefore, parameters have to be specified as follows:

V = <VOLUME> resels = <RESELS> Rd = <DLH>*<VOLUME>

• FSL smoothest: smoothness parameters can be calculated by the "smoothest" command line tool of FSL, as well. Output is to be handled as above.

• SPM: use the values from SPM.mat to rely on the SPM-like smoothness estimation:

  • search volume: SPM.xVol.S
  • number of resels: SPM.xVol.R(4)

Note that the pTFCE SPM Toolbox uses the values from the SPM.mat.

---

NOTE: the current implementation of the 4D residual-based smoothness estimation is not optimal and can be slow. Please consider using the FSL "smoothest" command line tool in case of processing time issues.

---