I believe this is handled by the 1000 Functional Connectomes project already. They’ve indicated that they dropped the first 5 scans from all sites data prior to releasing it, so the data that would have been downloaded and used for the Eklund et al analysis should have already had those first few scans trimmed off.
Mentioned here:
http://www.nitrc.org/forum/message.php?msg_id=6213
and here:
http://www.nitrc.org/docman/view.php/296/716/fcon_1000_Preprocessing.pdf
-Mike
--
Mike Angstadt
Research Computer Specialist / PANLab Lab Manager
Department of Psychiatry / University of Michigan
(734) 936-8229
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of John Ashburner
Sent: Thursday, July 21, 2016 2:00 PM
To: [log in to unmask]
Subject: Re: [SPM] cluster failure article
Hi Cyril,
I'm struggling with the comment system for the blog (possibly due to my version of FireFox). This is what I had intended to say:
In relation to the bit that says "some cases with the 1-sample t-test where the nonparametric approach had elevated FWE, due to skew in the data", it is important to stress that the first few scans of each fMRI run should not be included in the analyses. It is reasonable to guess that the skewing is due to these images being systematically different from those collected later in the run when a steady state has been reached.
All the best,
-John
On 21 July 2016 at 17:14, PERNET Cyril <[log in to unmask]> wrote:
Dear all,
There is now a blog post explaining in a different way the results the the Eklund et al. paper. We tried to be as didactic as possible
http://www.ohbmbrainmappingblog.com/blog/keep-calm-and-scan-on
Keep Calm and Scan On
www.ohbmbrainmappingblog.com
BY: JEANETTE MUMFORD, CYRIL PERNET, THOMAS YEO, LISA NICKERSON, NILS MUHLERT, NIKOLA STIKOV, RANDY GOLLUB, & OHBM COMMUNIATIONS COMMITTEE (IN CONSULTATION WITH THOMAS NICHOLS) In recent weeks a...
Cyril
________________________________________
From: SPM (Statistical Parametric Mapping) <[log in to unmask]> on behalf of PERNET Cyril <[log in to unmask]>
Sent: 19 July 2016 07:50:43
To: [log in to unmask]
Subject: Re: [SPM] cluster failure article
Yann,
The paper is quite clear, if you set p higher than 0.001 then your blobs becomes to big (under the null) for random field theory to be able to give you the right cluster size threshold and the nominal type 1 FWER is higher (ie you don't control as expected). This doesn't matter if its a one sample t-test or a contrast from an ANOVA, it is still wrong (see 1994 paper from K Friston).
Dr Cyril Pernet
Senior Academic Fellow
CCBS / Edinburgh imaging
Sent from my HTC mobile phone
----- Reply message -----
From: "Yann Quidé" <[log in to unmask]>
To: "[log in to unmask]" <[log in to unmask]>
Subject: [SPM] cluster failure article
Date: Tue, Jul 19, 2016 01:25
Hi all,
Just to jump on Mike's comment on wether an initial p=0.005 (+ cluster-wise correction) would be ok or not?
I understand there are less risks to report false positive blobs using p=0.001 as an initial threshold, and the need to (at least) use a strict p=0.001 (+ cluster-wise correction) when looking at within group activation and/or correlation. However, would it become a problem to use an initial threshold of p=0.005 (+ cluster-wise correction) when looking at, say for instance, 2x2 (between groups) ANCOVAs with clinical populations? This more liberal threshold will impact spatial sensitivity, but will it impact the validity of the findings?
Thanks.
Yann
On 14 Jul 2016, at 6:31 pm, Mike <[log in to unmask]> wrote:
> Thanks for everyone's replies. However, I believe that many researchers who use fMRI analysis software are not with a firm statistical background, just like me. For practical reasons, we need a "guideline," if any, to control multiple comparisons problem. Concerning cluster-wise thresholding, below is what I would follow according to Woo et al., 2004 and the recent cluster failure paper in PNAS, but I hope some erperts here can comment a bit.
>
> (1). For SPM and AFNI 3dClustSim users, the first arbitrary cluster-forming threshold (CFT) is suggested to be not too lenient. 0.001 is good, but 0.01 is definitely poor (I have no idea if 0.005 is ok or not?). Then you can report clusters that survive a FWE-corrected p<.05 at the cluster-wise level (but can I report FDR-corrected p<0.05?). The commonly used "P = 0.001 uncorrected with a k of 10 voxels" should be abandoned.
>
> (2). The commonly used "P = 0.001 uncorrected with a k of 10 voxels" should be abandoned (but it seems that many people still use it...).
>
> Besides, I have a naive question: since cluster-extent based thresholding might be more problematic, why don't we just stick on voxel-wise thresholding?
>
> Mike
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
**********************************************************
Electronic Mail is not secure, may not be read every day, and should not be used for urgent or sensitive issues
|