Hi John, cc List,
Just a quick thought on your remark; wouldn't it be problematic to
change data intensity scaling behavior in an spm5 update, and mixing
SPM5 subversions ina running study? For example people looking at
absolute parameter estimates (after unscaling them with the mean session
value found in the SPM.mat). We sometimes do this.I know that usually
SPM scales data before fitting a model by the session mean by default,
so this would only be a small issue.
Furthermore, some values would be differently affected by scaling or not
due to rounding, especially for very small statistical effects. Eg, when
you have the full bandwith available by a data type (say int16) for the
intensity range in your data, estimates might be slightly different when
compared with analyses done on an unscaled dataset where you only have a
fraction of your int16 bandwith available for the intensity range. This
would of course only be relevant for small-bitsize data and large
scaling factors in the nifti header (usally determined by the scanner
data output system).
You never know how a cow catches a rabbit, as we say in Dutch ;-)
In short: it might be wise to explicitely warn people not to mix
analyses between the current state of updates (958?) and future versions
affecting scaling? Then all would be groovy.
Or am I missing something crucial.
Cheers,
Bas
--------------------------------------------------
Dr. S.F.W. Neggers
Division of Brain Research
Rudolf Magnus Institute for Neuroscience
Utrecht University Medical Center
Visiting : Heidelberglaan 100, 3584 CX Utrecht
Room B.01.1.03
Mail : Huispost B.01.206, P.O. Box 85500
3508 GA Utrecht, the Netherlands
Tel : +31 (0)30 2509609
Fax : +31 (0)30 2505443
E-mail : [log in to unmask]
Web : http://www.fmri.nl/people/bas.html
--------------------------------------------------
-----Oorspronkelijk bericht-----
Van: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
Namens John Ashburner
Verzonden: maandag 26 november 2007 17:45
Aan: [log in to unmask]
Onderwerp: Re: [SPM] realignment with spm5 and with spm2
How do the values compare if the scalefactors in the image headers are
taken
into consideration? SPM5 automatically scales the images to use the
full
range allowed by that datatype (although this will change with the next
set
of updates - because of issues with 4D data), whereas (if I remember
correctly) SPM2 keeps the same intensity scaling as in the original
data.
Best regards,
-John
On Monday 26 November 2007 15:00, Sangkyun Lee wrote:
> Dear SPM users
>
> I am trying to realign fMRI data with SPM5 and SPM2.
>
> The result of realignment with SPM5 is different from the one of
> realignment with SPM2.
>
> Even though I set the quality of realignment same value in both the
> programs, the result was different.
>
> As I know, the algorithm of realignment in the SPM5 is the same as
> that in the SPM2.
>
> Is it right?
>
> If I use SPM2, the contrast of image is much larger than that in SPM5.
>
> For example, the value in the brain area is over 4000 in the image
> from SPM2, but is over 200 in the image from SPM5.
>
> What is the problem?
>
> Thank you!
>
>
>
> Best Regards
>
>
>
> Sangkyun
|