[Mristudio-users] LDDMM questions

[susumu mori]

Hi Yi,

1) I recommend you to do AIR first from two reasons. First one is a
practical reason. Our LDDMM requires that the matrix and pixel sizes of the
two images are identical. After AIR, even if the input images have different
dimensions, the output will have the same dimension. This is handy. Second
one is more important. Let me use fitting of a X-Y plot for analogy. When
you use linear fitting, no matter which software you use, the results are
almost identical. This is because the energy landscape of linear fitting
usually has a very simple "one-valley" shape, meaning the results always
converge at the same solution. On the other hand, when you are using
non-linear fitting, you are required to provide initial values, which are
close to the real solution. Usually you first do linear solution and then
use the result as the initial value for the subsequent non-linear fitting.
This is because the energy landscape of non-linear fitting has a complicated
"multi-valley" shape. If your initial values are far away from the real
solution, the fitting is trapped by a local minima.

Non-linear warping is very similar. You are required to bring two brains as
close as possible before you start non-linear warping.

When you adopt AIR+LDDMM, you have to transform the image twice, including
tensor reorientation. Landmarker can combine transformation matrices so that
you can do multiple transformation at once to avoid multiple interpolation.

Xin, can we combine AIR matrix and LDDMM matrix?

2) We usually don't use non-linear AIR. So we don't have much experience
with it. Yes, it should do tensor reorientation if you apply the
transformation to a tensor file.



On Mon, Jul 13, 2009 at 4:12 PM, Yi Jiang <yj3 at duke.edu> wrote:

>  Dear All,
>
> I have two questions about LDDMM:
>
>
>
> 1. if I have a template image and a subject image and I want to register
> these two by LDDMM eventually, 1). should I use LDDMM to register these two
> directly or 2). should I use AIR linear to affine transform them first, and
> then LDDMM the updated subject image? In the 2nd case, I need to reorient
> tensors twice, once by the AIR transformation matrix and once by the LDDMM
> transformation matrix, right?
>
>
>
> I am asking this question because for some other registration software I
> have used, it seems the elastic transformation works better after affine
> registering the images first.
>
>
>
> Or maybe LDDMM covers rigid and affine transformation well already so we
> can do #1?
>
>
>
> 2. How does the nonlinear AIR compensate/compare with LDDMM? Can the
> nonlinear AIR transformation be applied to reorient tensors?
>
>
>
> Thank you very much!
>
>
>
> Best,
>
> Yi
>
> _______________________________________________
> Mristudio-users mailing list
> Mristudio-users at mristudio.org
> http://lists.mristudio.org/mailman/listinfo/
> Unsubscribe, send a blank email to:
> Mristudio-users-unsubscribe at mristudio.org
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.mristudio.org/pipermail/mristudio-users/attachments/20090714/1a5a2cd6/attachment.html