Dear Carlos,
the new release r256 should solve that issue.
Extra question 1: The toolbox does not support any cluster-statistics and to be honest, I don't see the advantage or additional value over the TFCE statistics which is simply more powerful and sensitive.
Extra question 2: If your mask is too small the advantage of using the TFCE statistic might be diminished. Simply, compare TFCE and nonparametric T with the toolbox to make a decision.
Best,
Christian
On Thu, 24 Nov 2022 11:33:36 +0000, Carlos Murillo Ezcurra <[log in to unmask]> wrote:
>Dear Prof Christian Gaser / TFCE toolbox users,
>
>I am experiencing some problems with the implementation of TFCE in my SPM12 flexible factorial ANOVA (Factors: 80 subjects, 2 groups and 2 Timepoints) with an additional mask. The mask covers 10-15% of the whole brain. No nuisance covariates in the model. Thresholding mask = 0.2
>
>When I calculate TFCE in the whole brain analysis for T and F contrasts everything is ok
>When I calculate TFCE with the mask in my T contrasts everything is ok.
>
>However, when I calculate TFCE with the mask in my F contrast for main effect for time (defined as in the CAT12 manual), though results are estimated, I get a warning:
>“WARNING: Large discrepancy between parametric and non-parametric”
>
>I could not find yet the solution in previous related posts. I saw in previous posts that it was recommended to update the TFCE version to deal with this issue but when I update it to the latest one (r254), I got an error, and the model is not estimated:
>
>Error using spm_Tcdf (line 78) non-scalar args must match in size
>In file "/Users/carlosmurillo/spm12/spm_Tcdf.m" (v4182), function "spm_Tcdf" at line 78.
>In file "/Users/carlosmurillo/spm12/toolbox/TFCE/tfce_estimate_stat.m" (v234), function "tfce_estimate_stat" at line 832.
>
>Is there a solution for this issue? Can I trust the results with the large discrepancy warning from the TFCE older version or not?
>
>Extra question 1: Additionally, I am using TFCE for inference in the analysis of the study because its advantages on VBM data. However, I was wondering if it is also of added value to additionally report the parametric non-stationary cluster extent correction. As far as understand this could provide a different type of information from the TFCE correction.
>
>Extra question 2: I read in other posts that with small ROI's you usually recommend switching to voxel-wise inference because cluster size inference loses power. Should I consider voxel-wise inference with the size of my mask? Is there any study supporting voxel-wise inference for masked analysis?
>
>Thank you very much in advance for your time and consideration,
>Regards
>Carlos
|