Thanks, John.
I repeated the whole procedure on my linux box. Everything is OK
now. I was doing it on my SGI octane.
FYI:
I now know that the MATLAB running on SGI sometime generates wrong
numerical results if the math libray provided by SGI is not the right
version (now it is part of matlab shipment). To see if your matlab
Ok or not, try the following:
a=rand(90000,2); %90000 or a bigger number, but not terribly big
rank(a)
%if rank(a) is 2 and you get
0 0
0 0
when issuing b=a'*a
then your matlab has problems.
This is the case for my octane. I contacted mathWorks and am trying to
have it resolved.
(oh, my matlab version is 5.3.0.10183 (R11) on SGI64)
Any comments?
Kewei Chen
On Thu, 27 Jan 2000, John Ashburner wrote:
> Did you delete the .mat files of the original images before running
> the realignment a second time? The different starting estimates may
> produce tiny differences between the final results.
>
> Apart from this, I can't think of another explanation.
>
> All the best,
> -John
>
>
> | I want to run realign by-passing the spm-user interface (for possible
> | automation). First I use the spm interface (PET/SPECT).
> | realign
> | number of subjects: 1
> | then I selecte the two files (longitudinal studies from one
> | person)
> | Coregister & Reslice
> | Sinc Interpolation
> | All Images (1...n)
> |
> | The generated r*img (hdr, mat) files are then renamed as spmr*.img (hdr
> | mat),
> |
> | In running above, I find that
> | flagsC.quality=0.5
> | flagsC.fwhm = 8
> | flagsC.rtm =[]
> |
> | and
> |
> | FlagsR.hold=-9;
> | FlagsR.mask=1;
> | FlagsR.which=2;
> | FlagsR.mean=0;
> |
> |
> | then, I just run
> | spm('pet');
> | global MODALITY SWD
> | SWD=spm('Dir');
> | MODALITY='PET';
> | %the above codes may be un-necessary
> | spm_realign(P,flagsC);
> | spm_reslice(P,FlagsR);
> | where P is the 1X1 cell containing the two file names
> | to re-generate the r*img (hdr mat) files.
> |
> | I expect the spm-interface procedure generated r*img should be identical
> | to the one generated directly by running spm_realign.m and
spm_reslice.m.
> | What I found however, they are not EXACTLY identical. The maximue pixel
> | value difference, in most slices, is one (should be zero, if identical)
> | and in some slice levels, the difference is as high as 10.
> |
> | Visual examination of the two images could not detect any differences
and
> | pixel-by-pixel linear regression give almost a 45 degree line.
> |
> | Your help is greatly appreciated!
> |
> |
> | Kewei
> |
>
>
|