Dear Ning,
>Dear SPMers,
>
>I have some questions about realignment in SPM96.
>According to John Ashberner's suggestions given in last June (please see
>below), when I realign PET images in SPM96, I do it twice. But I am a bit
>confused by *.mat files generated in realignment procedure.
>
>When I do the 1st realignment, I choose:
>1) Coregister and reslice
>2) Sinc interpolation
>3) Mean images only
>4) Mask the images
>
>The results I get are: mean*01.img(hdr) and a set of *.mat files.
>
The contents of these files are:
1. mean*01.img(hdr) contains the image obtained if averaging all your
images, after having transformed the 2-n'th images according to the results
of the registration, i.e. according to the contents of the *.mat files.
2. *.mat contains the transformation matrix that will map each image file
onto the space defined by the 1st image. The transformation matrix is a 4x4
matrix defining three translations, three rotations and any differences in
image and pixel size between the first and the current image. The latter
probably don't apply in your case.
>Then I go to the 2nd realignment by using mean*01.img(hdr) and the
>original pet images *.img(hdr):
>1) Coregister and reslice
>2) Sinc interpolation
>3) All images + Mean image
>4) Mask the images
>
>This time the results I get are: meanmean*01.img(hdr), a set of
>r*.img(hdr) and a new set of *.mat files.
>
Here the contents of the mean*.* and the *.mat files are as above, only
that they reflect the results of the second registration (see below).
The r*.img files contain the transformed images, where the transformation
reflects the results of the second registration.
>
>In spm_realign.m, there is one sentence as follows:
>"Note that if the coregistration is performed more than once on the
>unresliced data, the starting estimates are obtained from parameters
>stored in the '*.mat' files."
>
>In my case, the coregistration is performed twice on the unresliced data
>(because the 2nd realignment also uses the original image sets), so the
>*.mat files obtained from the 2nd realignment are based on the *.mat files
>from the 1st realignment?? But the 2nd realignment also uses the original
>pet images with a reference (mean*01.img) different from the 1st one's
>(*01.img), why need the starting estimates of the 2nd realignment to be
>obtained from '*.mat' files generated by the 1st realignment?
>
First of all it should be realised that the orientation of mean*01.img
should (if all went well) differ very little from that of the first image
in the series since it is based on data realigned to match that.
Secondly, there are several reasons why it is a good idea to align images
to a "first pass mean image" rather than just the first image.
1. PET data are typically quite noisy and all registration methods are
negatively affected by noise. By realigning to a mean image rather than to
a single image the noise properties improve.
2. The realignment in SPM96 will perform 8 iterations, which will be
sufficient most of the time but may in the event of large movements be a
tad bit on the small side.
3. In SPM96 a small angle approximation (sin(a) ~ a) is used when
estimating the effects of rotations. Again, this is perfectly fine when
movements are small but may be a problem for larger movements.
2 and 3 both benefit from performing a second run of the realignment,
provided that the .mat files from the first realignment is used.
>For some reason, for some data set, before I do the 2nd realignment, I
>delete the '*.mat' files obtained from the 1st realignment, would this
>action affect the resulting r*.img(hdr) files and meanmean*01.img(hdr)??
>And how much is it do you think?
>
It would affect it, but it is difficult to say to which extent. I list
three reasons for the "two-stage" realignment above. By deleting the .m
files from the first pass you still have the benefits from point 1, but
loose the potential benefits from points 2 and 3.
My personal guess would be that provided that movements are in the range
one typically see for intrasession realignment (a couple of mm) you don't
need to worry even if you delete the .mat files from the first session.
>The generation of '*.mat' files is a new feature in SPM96. From SPM96 help
>I know that it is not necessary to get r*.img(hdr) images before the
>spatial normalisation stage. Could you tell me how to use '*.mat' files
>during the spatial normalisation? What should I pay attention to when I do
>this?
>
When reading *.img files SPM will automatically use the *.mat file with the
same name to transform the data. Hence, if you (or rather SPM) read a *.img
file together with the *.mat file resulting from the realignment or if you
read an r*.img file (which will have no associated *.mat file) the internal
results will be the same.
>From a practical perspective you needn't do anything special. Just select
your meanmean*.img to perform the normalisation and use the same parameters
to transform your *.img files. Just make sure you don't delete the second
set of *.mat files.
>Thanks a lot for your help.
>
Good luck Jesper
>On Thu, 18 Jun 1998, John Ashburner wrote:
>
>> Date: Wed, 17 Jun 1998 17:55:09 +0100
>> From: John Ashburner <[log in to unmask]>
>> To: [log in to unmask], [log in to unmask]
>> Subject: realigning PET images
>>
>> Personally, if using SPM96, I would realign everything to the first image,
>> and create a mean realigned image. This would be followed by a second pass
>> where all the original images are aligned to this mean. This could in
theory
>> be repeated many times, but I suspect that once is probably enough.
>>
>> SPM98 (SPM99?) will automatically use this two pass approach for PET
images.
>>
>> Regards,
>> -John
>>
>> > A quick question about realigning PET images. what is the best
technique?
>> > in the past i have selected 'all' scans and realigned in no particular
>> > order. i.e., the first scan selected for realignment was the first
listed
>> > alphabetically. is it better to produce a mean image and then realign to
>> > that, or is there a specific order in which i should select scans?
>>
>
>
>
>
>
>
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|