 |
 |
Yes, thanks for the clarification for the SPM lurkers on the FSL list :) On Tue, 2011-12-20 at 14:37 -0500, MCLAREN, Donald wrote: > The statement "FDR is strictly a voxel-based threshold" is not true in > all software packages. > > FDR is a statistical approach that leads to the conclusion that you > will have X % of the voxels are expected to be false positives. > > While it is true that in general FDR is applied at the voxel level and > that it is applied at the voxel level in FSL. In SPM, it is now > applied at the cluster level. > > See Chumbley and Friston. 2009. False discovery rate revisited: FDR > and topological inference using Gaussian random fields. NeuroImage. > Volume 44, Issue 1, 1 January 2009, Pages 62-70. > > Best Regards, Donald McLaren > ================= > D.G. McLaren, Ph.D. > Postdoctoral Research Fellow, GRECC, Bedford VA > Research Fellow, Department of Neurology, Massachusetts General > Hospital and > Harvard Medical School > Office: (773) 406-2464 > ===================== > This e-mail contains CONFIDENTIAL INFORMATION which may contain > PROTECTED > HEALTHCARE INFORMATION and may also be LEGALLY PRIVILEGED and which > is > intended only for the use of the individual or entity named above. If > the > reader of the e-mail is not the intended recipient or the employee or > agent > responsible for delivering it to the intended recipient, you are > hereby > notified that you are in possession of confidential and privileged > information. Any unauthorized use, disclosure, copying or the taking > of any > action in reliance on the contents of this information is strictly > prohibited and may be unlawful. If you have received this e-mail > unintentionally, please immediately notify the sender via telephone at > (773) > 406-2464 or email. > > > > On Tue, Dec 20, 2011 at 2:10 PM, Michael Harms > <[log in to unmask]> wrote: > Setting q=0.01 in FDR returns the p-value threshold such that > 1% of the > voxels below that threshold (i.e. above .999972559 in your > case) can be > expected to be false positives. FDR is strictly a voxel-based > threshold, and the clustering or spatial organization of the > p-values in > your "p1" volume has no impact on the p-value threshold that > is > returned. Thus, FDR and Gaussian Random Field cluster-based > correction > (i.e,. FEAT) are two very different things. > > cheers, > -MH > > within the On Tue, 2011-12-20 at 16:37 +0000, Dar Meshi wrote: > > Hello, > > > > I conducted an fMRI experiment and performed a contrast > using FEAT cluster correction for multiple comparisons, > setting the z-threshold to 2.3 and the p-value to 0.05. This > analysis yielded a nice result, with large clusters (around > 400-600 voxels). However, when I did the exact same contrast, > correcting for multiple comparisons with FDR and setting q to > 0.01, the clusters are much smaller (around 20 voxels), with > the peak voxels in the same location. Would you happen to know > why this is? I may be making a simple calculation error, but I > believe the z-threshold of 2.3 in cluster correction should > the same as a q of 0.01 in the FDR analysis, right? And the > FDR should be less conservative than the cluster correction in > FEAT, right? > > > > A couple notes: > > > > 1. I used outlier de-weighting in my FEAT analysis > > 2. I performed the following 4 command line steps for the > FDR analysis: > > ttologp -logpout logp1 varcope1 cope1 `cat dof` > > fslmaths logp1 -exp p1 > > fdr -i p1 -m ../mask -q 0.01 > > fslmaths p1 -mul -1 -add 1 -thr .999972559 -mas ../mask > thresh_1_minus_p1 > > > > You can see that the number I get from the "fdr" step is > quite small. Any chance you know what's going on? > > > > Thanks so much and happy holidays! > > Dar > >