> We are interested in using T1 normalization parameters to normalize our
> functional images during preprocessing.
>
> Currently our preprocessing looks like this:
>
> -"Display" the first functional image and reorient it to the AC-PC line
Ideally, you would reorient all the functional images the same way, rather
than just the first. This is done by selecting all the fMRI, rather than
just the first one, after clicking "Reorient images...".
> -"Realign"; select all the functional images; coregister only
Yes.
> -"Normalize"; determine all parameters and write; template = EPI.mnc;
> source image = first functional image; images to write = all functional
> images -"Smooth"; 8mm; select all normalized functional images
That should do it.
>
> Our proposed method of using T1 normalization parameters is this:
>
> -"Display" first functional image and reorient it to the AC-PC line
> -"Display" subject's T1 image and reorient it to the AC-PC line
> -"Realign"; select all the functional images; coregister and reslice;
> create mean image only
> -"Coregister"; coregister only; target = mean EPI; source = subject's T1
> image; other images = none
> -"Normalize"; template = T1.mnc; source = subject's T1 image; images to
> write = all functional images
> -"Smooth"; 8 mm; select all normalized functional images
> Is this an acceptable method? Are there any downsides/dangers to using
> this method?
The problem is that it probably doesn't work so well. Spatial normalisation
in SPM2 parameterises the deformations by only about 1000 parameters, which
is not enough degrees of freedom to obtain a really good model of shape
variability. Because matching is done with images that are not
skull-stripped, many of the degrees of freedom go towards making the scalp,
face and neck match the template. If the data are skull-stripped, and the
template is also skull-stripped, then a more accurate result could be
obtained. This is what I used for the evaluations done in a paper by Hellier
et al.
If you use SPM5, then you should find that better spatial normalisation can be
achieved using the Segment button. If you are still using SPM2, then maybe
you could try the so called "optimised VBM" approach, which involves
segmenting the T1 and warping the grey matter to match a grey matter tissue
probability map template. When registration is done this way, there is no
scalp, face or neck to fit. All the DOF go towards aligning the GM, which is
generally what you want for fMRI studies.
Another issue is whether the fMRI is distorted in the phase encode direction.
If you have arelatively old scanner, then you may find that the results of
the coregistration are not that great. This is because coregistration uses a
rigid-body model, which does not correct the distortion artifacts. These
could be corrected by field maps, using Chloe Hutton's FieldMap toolbox and
Jesper Andersson's "Realign and Unwarp" option.
Best regards,
-John
|