alignment, registration and motion correction

[tirzan]

dear all,
I'm dealing with fMRI time series (a sequence of not-aligned low-
resolution images, affected by artefacts due mainly to the patient
motions).
The goal is to align the images one into the others (usually all the
images are aligned to the first one).
My problem now is to have a clear idea of the therms: "alignment",
"registration", "motion correction", to me they seem synonymouses,
but
they are not.
"registration" is a synonymous of "motion correction"?
What "motion correction" is in your opinion?

I got from pixel.to.life the following definitions:
"alignment": To overlay two images in a way that maximizes the
combined information conveyed when the overlayed images are visually
perceived.
"registration": To estimate an appropriate transformation between two
given images that allow one to perform "alignment" as described
above.
"motion correction": To estimate an appropriate transformation method
that would allow one to "register" two given images with a clear goal
of reducing the mis-alignment of objects of interest solely due to
both global (rigid, e.g. skull tilt, translation etc.) and local
(non-
rigid, e.g. soft tissue movements inside the brain matter) motions.

but I already have not clear few things:

alignment is = registration + interpolation ?
"registraton" and "motion correction" still seems to me synonymouses.
Probabily the difference is that motion correction tries to reduce
mis-
registration (false positives + true negatives) and the registration
is simply the transformation. Is it plausible?

thanks a lot,
tirzan

[jme...@uci.edu]

Hi Tirzan,

Here is another definition of the terms:

"registration" - overlay of two images using corresponding points
or elements for comparison (manual or automatic)
"alignment" - linear or non-linear deformation of an image to match
a reference image (manual or automatic)

(Caution: In the literature, these terms are often used
interchangably.)

"motion correction" - correction in a temporal or spatial series of
images
for motion of the patient during image acquisition (usually a linear
transformation)

Best regards,

Joerg, the Scienctific Medical Imaging Enthusiast

[Pixel.to.life]

On Apr 16, 12:36 pm, "tirzan" <tirza...@gmail.com> wrote:

Hi, Tirzan,

As you know too, to visually assess the result of a registration
process, there has to be a 're-sampling' of the subject image
employing the result transformation.
Interpolation would then come into play if the resulting transform
does not overlay the two images at absolute grid locations. So yes, in
that sense, to "align" two images using a "registration" result, we
have to "resample" the subject image.

Also, to my knowledge, one employs "registration" trying to compensate
for (or to reverse) motion of structures between the instances that
the two images were taken. Hence, the goal being "motion correction".
The method to get the transformation being "registration", the method
to apply this transformation being "resampling" (involving
interpolation).

Interesting exercise it has become:-)