 |
 |
Hi, Anderson Thanks for your kind help! I have three more questions regard this problem. You say that TFCE is generally more powerful. What does this mean? The TFCE method is more reliable? or If I use both the TFCE and the cluster method, there are more likely survived clusters in the _tfce_corrp_tstat file rather than the _clustere_corrp_tstat file? If no voxel can pass the TFCE FWE correction, could I use the cluster-wise correction method? If so, how could I set the cluster-forming threshold? I need set the T threshold, for eg, 3 or 4, right? Could I set the p value, such as p<0.001? Because I know the degree of freedom and the p value, so I can compute the corresponding t value using matlab function tcdf and tinv. Could I? Finally, I find that when I finish randomise, it displays "Critical Value for: tbss_clustere_corrp_tstat2 is: 279; I think this might be the cluster size I need. It may mean that combining a cluster-forming threshold T = 3 and cluster size of 279 could achieve FWE p<0.05 cluster-wise corrected, right? Best, Feng 发件人："Anderson M. Winkler" <[log in to unmask]> 发送时间：2015-11-04 16:36 主题：Re: [FSL] Questions about the Cluster-wise correction in FSL randomise 收件人："FSL"<[log in to unmask]> 抄送： Hi Feng, Please see below: On 3 November 2015 at 14:14, chenhf_uestc <[log in to unmask]> wrote: Dear FSL experts, I have several questions about the Cluster-wise correction in FSL randomise, in the manual (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Randomise/UserGuide). First, there are two cluster-wise method: the Extent and Mass. The manual states that 1. Cluster-based thresholding corrected for multiple comparisons by using the null distribution of the max (across the image) cluster size (so passé!): <output>_clustere_corrp_tstat / <output>_clustere_corrp_fstat.; 2. Cluster-based thresholding corrected for multiple comparisons by using the null distribution of the max (across the image) cluster mass: <output>_clusterm_corrp_tstat / <output>_clusterm_corrp_fstat. What's the difference betwwen these two methods? Is there any reference that introduce there two methods? The cluster extent is the number of voxels in a cluster, whereas the mass is the sum of the statistics within a cluster. I think the original references are somewhat scattered. Cluster extent appeared in a paper by Friston and Poline among others I think in 1994, then in another paper by Forman in 1995. Cluster mass appeared in Bullmore et al 1999. I think a concise description may be available in the Poldrack/Mumford/Nichols book: http://www.amazon.co.uk/Handbook-Functional-MRI-Data-Analysis/dp/0521517664 Second, the cluster level correction usually need a voxel-wise threshold for forming the cluster (in the SPM or Freesurfer software, I can set voxel p<0.01), but here, in the FSL, the manual states use -c <thresh> for t contrasts, where the threshold is used to form supra-threshold clusters of voxels. Does this mean that I need set a T value? Yes. Third, Could I use cluster-wise correction in the TBSS results? Or I just can use the voxel-wise FWE corrected and TFCE based FWE corrected method? If I can use cluster-wise correction, which correction method is more easily to pass? (The TFCE or the cluster-wise? The voxel-wise FWE method, in my experience, is difficult to pass). TFCE is generally more powerful. Finally, in the SPM, if I use the cluster level correction. For e.g., I use an uncorrected voxel p<0.01, SPM will tell me that I need set a extent cluster size threshold of "some" voxels. Thus, the uncorrected voxel p<0.01 and the reported cluster size will reach the corrected cluster level of p<0.05. But in the randomise, I do not know where can find the cluster size threshold. There is no such a thing in randomise. All the best, Anderson Any help will be great appreciate. Best, Feng 2015-11-03 chenhf_uestc