I'm not sure what the current state-of-the-art is, but you could maybe look
through some of the papers that cite this one...
Title: Positron emission tomography metabolic data corrected for cortical
atrophy using magnetic resonance imaging
Author(s): Labbe, C; Froment, JC; Kennedy, A, et al.
Source: ALZHEIMER DISEASE & ASSOCIATED DISORDERS Volume: 10 Issue: 3
Pages: 141-170 Published: FAL 1996
Your procedure sounds roughly OK, although there probably isn't a need to
binarise the tissue class images. Also, if you coregister first, you could
use the *_seg_sn.mat from the segmentation (of SPM5) to do the spatial
normalisation.
In response to your questions:
1) If you use SPM5, then spatial normalisation and tissue classification are
combined in the same model. It warps tissue probability maps (generated from
the average tissue class images of lots of subjects) so that they are
overlayed on to the original image. The inverse of this transform can be
used for doing the spatial normalisation.
2) You will need to do some sort of masking to avoid divisions by very small
numbers. If you plan an SPM analysis of spatially normalised data, then
things may become a bit more complicated, especially if you plan to smooth
your spatially normalised data. I may be wrong, but if it was my data, I
would process it by:
* Coregister SPECT and MR.
* Segment the MR to obtain GM and WM in native space.
* Smooth the GM and WM by the PSF (this may be easier sad than done if it is
not isotropic - in which case you may need to do some sort of reorientation
of the tissue class images).
* Warp the SPECT and the smoothed GM and WM according to the *_seg_sn.mat file
generated from the segmentation - making use of the preserve total option
(which applies a Jacobian transform, or "modulation").
* Apply the usual additional smoothing of the warped data (8-12mm??).
* Subtract the smoothed, modulated, warped, smoothed white matter (scaled by
some value) from the smoothed, modulated, warped SPECT, and divide this by
the smoothed, modulated, warped, smoothed grey matter.
* Mask out regions where the smoothed, modulated, warped, smoothed grey matter
has a value of less than about 0.05 - or use some form of explicit mask for
the SPM analysis.
Best regards,
-John
On Thursday 26 June 2008 03:26, Emma Schofield wrote:
> Hello,
> I have SPECT and MR scans from controls and patients, and I want to run a
> group comparison of the SPECT scans.
> I am trying to perform some kind of atrophy correction on the SPECT scans
> before running the statistics and think that the following method would be
> the best way.
>
> Coregister raw SPECT and raw MR images (with reslicing to bring them into
> line) Segment coregistered MR images (without prior normalisation so that
> the atrophied regions are not warped)
> Take the GM image and make a binary mask
> Take the WM image, and make all the WM pixels have an average WM value
> (measured using an ROI)
> Smooth the binarised GM and WM images to the point spread function of the
> SPECT imaging system.
> Subtract the WM image from the SPECT image (assuming no HMPAO is in the
> CSF) Divide the SPECT image by the GM smoothed binary image
> Normalise corrected SPECT scans to SPM5 SPECT template
>
> I would appreciate any comments on the effectiveness or technicalities of
> this process. I also have a couple of other questions,
>
> Firstly, is there any problem with segmenting without prior normalisation?
> (It seems that segmentation is possible without it, and I don’t want to
> warp the MR at all before the atrophy is corrected for).
>
> Secondly, should I mask the final resultant image with the unsmoothed GM
> image?
>
> Thanks very much :-)
>
> Emma
|