 |
 |
Hi Anderson, Thank you for answering me, but I didn’t really manage to find what « could be done » and what is « mathematically true with no possible refutation » concerning TFCE. To explain you in details, I made a study on 10 mm FWHM on both surface and volume metabolism maps using corrected (FWE p < 0.05) TFCE statistics, with PET scanner having a 5.7 mm spatial resolution. I chose a minimum cluster size of 90 mm2/mm3. Below the question from the reviewer: « how was the size of the accepted cluster chosen? 90 mm2 considering the actual spatial resolution of the used scanner and the filtering at 10mm is very "dangerous" and possibly unreliable. As it is 90mm3 in the subcortical structures if I interpreter well "resp.", about 4.5x4.5x4.5 mm, below the camera resolution » Could you give me your opinion on this ? Best regards, Matthieu > Le 10 juin 2017 à 16:07, Anderson M. Winkler <[log in to unmask]> a écrit : > > Hi Matthieu, > > There is no need for a minimum cluster size because for cluster, the question is "if the null is true, what is the probability of finding a cluster at least as large as this one I found with my statistical test". Putting a minimum cluster size will not change that probability, as the even the clusters that are under the threshold would pile up on the left side of the permutation distribution of the maximum cluster size, as needed for the FWER correction. > > And If the cluster found with the test is smaller than the desired threshold, then the permutation test wouldn't even need to be done, such that including a minimum cluster size before doing the test is equivalent to deleting these small clusters at the end, that is, it's an arbitrary decision. > > All the best, > > Anderson > > > On 7 June 2017 at 13:20, Matthieu Vanhoutte <[log in to unmask] <mailto:[log in to unmask]>> wrote: > Dear PALM's experts, > > Having done volume and surface-based metabolism analyses with TFCE statistics (PALM), I wonder if as other neuroimaging statistical tools it is needed to threshold corrected p-maps by minimum cluster size ? > > I could see through mailing list that "if corrected TFCE results are significant, in spite of being small, they should be considered as such". If it is the case, could you explain me why and address me a paper reference ? > > Best regards, > Matthieu >