Dear Claus,
I see...I thought that choosing "none" and entering T=3 resulted on a
FDR-corrected SPM. I misunderstood Guillaume's explanation.
Now I understand, but my results don't show any expected cluster of interest
with the proper FDR threshold...
The fact is that I'm working with images from a Siemens 1.5 T scanner, and
its analysis software shows more activated areas than this corrected SPM
(with a lot of noise also). I know it uses General Linear Model to analyse
the data, but I don't know if it shows corrected or uncorrected results. I
would like to see if SPM can get clearer and reliable results.
I read something about a difference between topological and voxel-wise FDR,
and the possibility of configuring SPM8 to use the latter editting
spm_defaults.m so that, l.42: defaults.stats.topoFDR = 0;
Is it possible to use this voxel-wise FDR to get more desirable (more
clusters) corrected results?
Thank you so much for sharing your knowledge
Álvaro
On Wed, 17 Mar 2010 17:26:32 +0100, Claus Lamm <[log in to unmask]> wrote:
>Dear Alvaro,
>
>Not sure whether this is a misunderstanding from my side, but if you enter T
>= 3
>after having clicked "None", you will get an SPM thresholded at an
>uncorrected T-value of 3
>(which in my experience, depending upon dfs of course, is usually equivalent
>to P-values around 0.001)
>
>So this quite liberal threshold might explain why you get "expected
>activations" rather than with the more
>conservative FDR or FWE threshold - at the cost of not having control over
>type I errors, though.
>
>Note, therefore (unless I misunderstand your question), that the T=3 rule of
>thumb is only used to obtain the FWEc and FDRp/c,
>but should not be mistaken as the actual corrected threshold.
>
>Hope this helps
>Claus
>
>-----Original Message-----
>From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On
>Behalf Of Alvaro Muro
>Sent: Mittwoch, 17. März 2010 15:52
>To: [log in to unmask]
>Subject: Re: [SPM] FDR p-threshold in SPM8
>
>Dear Claus,
>
>Thank you very much for answering so fast.
>
>For the first Results procedure, what method/threshold is usually chosen?
>I've noticed that those four values vary depending on the option selected
>the first time.
>
>Moreover, on your conversation with Guillaume he says:
> "For topological FDR, I would typically choose "none" and a small T-value,
>for example 3."
>
>In my current study, the FDRp shown is 5.808, and it seems too conservative.
>When I use "none" and "T=3" I get the expected activations. Is it possible
>to know the equivalent p for that T=3 threshold? Is this T=3 like a standard
>value for FDR correction, as p<0.05 is for FWE?
>
>Thanks for your help!
>
>Álvaro
|