On Tue, 22 Apr 2008 00:38:50 +0100, Vivien Tsen <[log in to unmask]>
wrote:
>Based on previous posts, the advantage of using structural (T1-weighted)
>image for co-registration and spatial-normalization as opposed to
>normalizing functional images directly to the EPI template is that:
> 1. T1-weighted images have higher spatial resolution which should
allow
>better overlap with the corresponding structures in the template.
> 2. there is the issue of EPI distortion variability in subjects and
>that of the template (which would be less of an issue in structurals).
>
>My question is:
>if the EPI images are prone to distortion, it is quite likely that the
>functional image (EPI) and the structural image (T1-weighted) cannot be
>matched by rigid-body transformation (realignment or co-registration)
>alone. So, if the EPI cannot be well-matched with the structural, how
much
>of an advantage is it using the T1-weighted for unified segmentation
after
>co-registration? Has a comparison been published between the two methods?
I don't know about anything having been published, but this is a very
interesting question.
Your point about EPI/T1 matching is a good one. Note that there's no need
for EPI distortion to be zero; we only need that the distortion is small
enough that the resulting errors in coregistration (and ultimately mapping
into canonical space by the following nonlinear warping) is reasonably
small. (You allude to that with your phrasing "well-matched".)
One time I looked in more detail at coregistration, in the context of
comparing different cost functions (this was SPM99, I think). The
coregistration was making a pretty good match, to the eyeball, but it was
introducing a rotation on the order of 3 degrees, when in fact from the
motion data and other considerations I thought that was excessive. So it
occurred to me that even reasonable amounts of EPI distortion could lead
to pretty large errors in inter-modal coregistration. But it's an
empirical question, really, and I haven't spent any time pursuing that.
In principle this could be finessed by using an unwarping toolbox, though
I don't know how good the EPI unwarping algorithms are.
If one really knew that the subject hadn't moved much between the EPI and
T1, one could just use scanner coordinates to match them.
>
>Thanks.
>========================================================================
--------------------------------------------------------------------------
CONFIDENTIALITY NOTICE: This e-mail message, including any
attachments, is for the sole use of the intended
recipient(s) and may contain confidential, proprietary,
and/or privileged information protected by law. If you are
not the intended recipient, you may not use, copy, or
distribute this e-mail message or its attachments. If you
believe you have received this e-mail message in error,
please contact the sender by reply e-mail and destroy all
copies of the original message
--------------------------------------------------------------------------
CONFIDENTIALITY NOTICE: This e-mail message, including any
attachments, is for the sole use of the intended
recipient(s) and may contain confidential, proprietary,
and/or privileged information protected by law. If you are
not the intended recipient, you may not use, copy, or
distribute this e-mail message or its attachments. If you
believe you have received this e-mail message in error,
please contact the sender by reply e-mail and destroy all
copies of the original message
|