[Insight-users] EPI Ground Truth

Arash Jahangir arash at vije.ca
Thu Dec 22 08:50:46 EST 2005


Hi Martin,

> I am not sure what is the aim of your work, but...I would chose
> other approach to evaluate performance of a non-rigid algorithm.
> An object (phantom) with known shape and known deformation, imaged with
> modality which doesn't suffer from deformation, such as CT.

The object of known shape is the conventional T2 brain scan.  The deformed 
object to be registered to it is the same T2 scan that has been deformed to 
resemble an EPI image.  The registration schemes will be judged based on how 
closely they can reproduce the original T2 scan given either {Original-T2, 
Simulated-EPI} or {Field map measurements,  Simulated-EPI}.  The first 
dataset will be of use for those who wish to unwarp/register the epi in 
spatial domain, the latter for those who want to correct in k-space.

Either way, the problem is (1) how realistic does the simulated epi need to 
be so that its registration with the original T2 is a good measure of 
performance of the registration scheme?  (2) How can such realism be 
achieved.

Since I expect a fair chunk of ITK community to use the results of the final 
test schemes, I am looking forward to their feedbacks on above issues.

Note that I don't have a problem in actual procedure of deforming the 
original T2 image.  My issue is how realistic do the contrast/intensity 
range of the simulated EPI have to be and how realistic the deformations 
themselves have to be.  For example, if I have one set of registered epi-t2 
images, is it good enough to deform another T2 image based on the warp 
produced from the first set in order to produce a perfect T2-EPI set?

For those doing the post processing in spatial domain, if they are using 
some intensity invariant metric, say normalized mutual information, then 
perhaps recreating the actual intensity/contrast ranges does not need to be 
so exacting.

For those who are correcting the epi in k-space, then I suspect a more 
exacting measure is needed, but I am not sure.

what do you think?

Arash 



More information about the Insight-users mailing list