We have found that using the FA gives
better registration results than the b0 when registering to T1 images.
Also, if you have EPI DTI images, you may not be able to get perfect
intrasubject registration with rigid body registration (or even affine) because
of non-liner susceptibility warping of the EPI DTI images. If you really
need exact registration, you will probably have to correct the distortion in
the DTI images.
Peace,
Matt.
From: FSL - FMRIB's
Software Library [mailto:[log in to unmask]] On
Behalf Of Jiansong Xu
Sent: Saturday, January 19, 2008
1:32 PM
To: [log in to unmask]
Subject: Re: [FSL] Flirt--
coregistration between T1 and DTI B0 image
Thanks a lot.
Jiansong
On Jan 19, 2008, at 1:07 PM, Ted Yanagihara wrote:
Hi Jiansong,
So that is one way to do it, but what I was suggesting was to register to your
lowres brain (i.e. example_func) to your subject's T1, then invert that matrix.
You would be drawing your masks on the T1 and then applying your inverted
matrix to the masks.
I guess I made it more complicated than it needed to be, because for your
purposes you could have used the 2-step registration (lowres->highres) and
left the MNI space out of the equation. I always do the 3-step because it gives
better registration to standard space when doing a group analysis, so I just
create all my transformation matrices from the start.
For the bigger picture, this is all a nice way to "quickly" check and
see what a subjects FA values are for your ROIs, but you should follow Saad's
advice about using TBSS. I believe this is the real analysis you are looking
for and does not require ROIs.
Good luck!
ted
On Jan 19, 2008 12:05 PM, Jiansong Xu <[log in to unmask]> wrote:
Thanks a lot.
So, this approach is to normalize every subject into standard space,
and draw ROIs in standard space. Then, use inverted matrix of each
subject to transform the ROI in standard space into the native space of each
subject. Am I right?
Best
Jiansong
On Jan 19, 2008, at 11:24 AM, Ted Yanagihara wrote:
I think I may know what
will work, but in the end you can check and see if this is what you are looking
for.
Use the flirt with three steps (low->high->ref). Use Affine (12 parameter
model) to register the high res to the reference. This is because you need to
use all the degrees of freedom to transform your subject's highres image into
reference space (these are very different brains). Use the Rigid body (6
parameter model) for the lowres to highres transformation (they only need a
little shifting since they are the same brain). Once this is complete, you will
have 3 transformation matrices. I believe they are numbered in the following
way: low2ref.mat = lowres to standard; low2ref1.mat = highres to standard;
low2ref2.mat = lowres to highres. Now you have all the transformation matrices
that you should need for the rest of your analyses.
Take the matrix that transforms from lowres to highres space and invert it
using invert flirt transform in the utilities. Now you want to apply this
matrix, which is highres to lowres, to your T1. You may actually want to draw
your masks in highres space and then apply the inverted matrix to the masks
instead. See what works for you though.
I am pretty sure this is what I have been doing and it has been working with
good results.
ted
On Jan 19, 2008 10:39 AM, Jiansong Xu < [log in to unmask]>
wrote:
Dear Friends:
I'm trying to define ROIs during analysis of DTI data, and find that
the resolution of both B0 and FA images are too low. I want to co-
register T1 and B0 and FA map, so that I can take the advantage of
high resolution of T1 image and define ROI on T1 image of each subject.
I used Flirt with Rigid Body (3 parameter model) to co-register the T1
with either B0 or FA map of the same subject, but the result was not
good. I'm wondering if anyone can make suggestions.
Best
Jiansong