 |
 |
Thanks Feng! Yann On 23 Oct 2015, at 12:46 pm, chenhf_uestc <[log in to unmask]> wrote: > Hi, Yann > > This result is right and reasonable. In the VBM analysis, the smooth is not uniform. In smooth regions, clusters tend to be large even in the absence of true signals, thus resulting in increased false positives. On the other hand, in rough regions, clusters tend to be small, and even a true positive cluster may be too small to be detected, resulting in reduced power. The non-stationary method actually adjusts cluster sizes according to local roughness of images. > > So, the cluster that has 2178 voxels in not significant (that means, you know, this area is more smoother) > > For more information, see reference "Nonstationary cluster-size inference with random field and permutation methods". > > Hope this helps. > > Best, > Feng > > 发件人：Yann Quidé <[log in to unmask]> > 发送时间：2015-10-23 09:19 > 主题：[SPM] Question regarding cluster FWE correction for VBM8 study > 收件人："SPM"<[log in to unmask]> > 抄送： > > Dear all, > > I have a question regarding cluster FWE/FDR correction for F-test (3x2 between group ANOVA, with covariates) in SPM12 following VBM8 analysis (I used SPM12 only to get the cluster correction stats, everything else has been processed and analysed using VBM8/SPM8; the peak-level stats are identical in SPM8 and SPM12). In spm_defaults.m, I changed the defaults.stats.rft.nonstat = 0 to 1 (line79). > > > As you will see attached, I have surprising pattern of statistics. I used an initial uncorrected threshold of p=0.001, and wanted to know what was the FWE/FDR appropriate cluster extent to take into account the multiple comparisons. At the end of the table, SPM12 gives me a value of 1200 voxels for both FWE/FDR. However, when I look at the stats for the clusters that are bigger than 1200 voxels, SPM12 reports 1 cluster made of 2178 voxels, with a FWEc value of 0.087 (ie >0.05)! > > How is that possible, and how this should be handled? Is this cluster really not significant? What is better to report in that case? > > Any idea? > > Thanks for your help. > > Yann > > > <Screen Shot 2015-10-23 at 12.02.05 pm.png>