Mismatch Between xjView and SPM GUI in Presentation of Results

[Alon Erdman]

Hello,

I have been using the SPM program for fMRI analysis for a long time and recently I have noticed there is a serious mismatch between xjview and SPM GUI when it comes to the presentation of results after FDR/FWE correction. While some contrast "survives" FDR correction in xjview, the same contrast isn't deemed significant when opened in the SPM GUI.

I'm aware that the default in the SPM GUI is FWE correction while in xjview it's FDR, but the gap between them is huge, and from my knowledge it's not supposed to be like this, and even if there is some gap- the FDR correction should be more severe and I have found the opposite. 

I thought that maybe I made a mistake, but I checked it out with two other researchers in my lab, about separate data, and they found the same problem.

It's important to emphasize that this exists only when it comes to FDR/FWE corrections, and not in regular p value statistics. 

I would appreciate any guidance that can help me understand this problematic situation.

[Xu Cui]

Alon,

Thanks for sharing your feedback. We have not compared fwe and fdr before. Do you mind sending your images and screenshots?

Xu

--
You received this message because you are subscribed to the Google Groups "xjView discuss group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to xjview-discuss-g...@googlegroups.com.
To post to this group, send email to xjview-dis...@googlegroups.com.
Visit this group at https://groups.google.com/group/xjview-discuss-group.
For more options, visit https://groups.google.com/d/optout.

--
Sent from Gmail Mobile

[Rich Morris]

I can confirm a discrepancy exists between the FDR-corrected voxel level p-values. This is present in the current version (8.14) and earlier versions (8.12). There doesn't seem to be any discrepancy between FWE or uncorrected results (either at the voxel level or cluster level).

-Rich

[Rich Morris]

Update: Mismatch appears to be due to the newer versions of SPM (SPM8, SPM12) which provide FDR control by topological properties of the signal, superseding earlier versions which provided FDR by voxel. If you set defaults.stats.topoFDR = 0; in spm_defaults.m, then SPM returns FDR values which match xjView. See here for details: http://www.nemotos.net/?p=139

Hope that helps,

Rich

On Thursday, March 24, 2016 at 1:33:29 AM UTC+11, Alon Erdman wrote: