Cornelius,
Do you actually get vastly different z-values, or simply differences in the
thresholded blobs showing up after correction for multiple comparisons? I
believe that the correction for multiple comparisons work to different
criteria between SPM2 and FSL. From my limited knowledge of FSL (peaking
over my colleague's shoulder), it seems that FSL considers a voxelwise
z-score of around > 2.7 as "significant" (corrected), whereas in SPM2, at
typical smoothing kernel sizes, a z-score of around 2.7 would not even make
it to uncorrected p 0.001.
(I might of course be wrong about this, and if I am it would be interesting
to hear about the differences in multiple comparison correction between the
two programs...is the FSL correction really more lenient than the SPM2
correction?).
beswt wishes,
Tobias.
Tobias Egner, PhD
fMRI Research Center
Columbia University
Neurological Institute, Box 108
710 West 168th Street
New York, NY 10032
Tel: (+1) 212 342 0121
Fax: (+1) 212 342 0855
Email: [log in to unmask]
> -----Original Message-----
> From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On
> Behalf Of Cornelius Werner
> Sent: Monday, December 06, 2004 10:07 AM
> To: [log in to unmask]
> Subject: Re: [SPM] [FSL] SPM2 and FSL 3.2b
>
> On Mon, 06 Dec 2004 14:23:54 +0100, John Ashburner
> <[log in to unmask]> wrote:
>
> (cut)
> > Because the images are in the SPM format,
> > rather than the FSL format, you would get blobs in SPM, but not FSL.
> (/cut)
>
> Interestingly it was FSL that gave me blobs where SPM wouldn't. I hope
> that doesn't ruin my credits with the SPM community :-)
> I'll try and track it down - I still believe it has to be a mess-up on my
> part.
> Best regards,
> Cornelius Werner
>
>
>
> >> just out of curiosity, I analyzed a fairly complex (and long) erfMRI
> >> experiment with a 3x3 factorial design with both SPM2 and FSL3.2b. I
> >> tried
> >> to keep everything as close to the other package as possible, i.e. I
> >> adjusted both smoothing kernels to 5mm, set the high pass filter to
> >> 128s,
> >> used gamma function + temp. derivative in FSL (standard delay) /
> >> canonical
> >> hrf + tdv in SPM2, entered identical onset vectors, specified identical
> >> contrasts and so on. I set the duration for my events to the actual 1.5
> >> seconds in both packages, as FSL doesn't allow 0-durations.
> >> Now one of the packages gave me "blobs" on a
> >> corrected-for-multiple-comparisons level of 0.05, where the other
> didn't
> >> show anything worth mentioning on a uncorrected p=0.001 level. Is that
> >> possible at all, given the similarity of the statistics? Has anyone
> >> experienced something similar? I am aware of the comparison done by
> >> Bianciardi et al (NeuroImage 2004), but differences were marginal at
> >> best...!
> >
>
>
>
> --
> Cornelius Werner
> Institut fuer Medizin (IME)
> AG Kognitive Neurologie
> Forschungszentrum Juelich
> 52425 Juelich
> Germany
>
> Tel. +49-(0)2461-61-8609
|