Hi,
As you found out, flirt doesn't care about the sign of the pixdim
entries
when registering images. What the pixdim is used for is for assigning
"talairach/mni" coordinates to voxels. In this case the negative sign
does
have a meaning, and determines whether a direction has increasing
or decreasing coordinate values as along the voxel coordinates.
All flirt does internally, is take the data and the absolute sizes for
the
voxel dimensions and then register these images.
Note that whether something is displayed one way or another is a
separate issue determined by the display program.
Generally absolute anatomical references (ie. RAS or LAS) are avoided in
all fsl programs as often the conversion to Analyze does not retain
sufficient information to determine such things. The only case where
anatomical information exists is in the standard images, and
particularly
the talairach/mni coordinates associated with them. We only use the
avg152 image and its associated header (including the negative pixdim)
to determine these coordinates. How the various programs behave for
other images with negative pixdim and non-zero origins is not something
we have tested and may not be consistent. We assume that all other
images have positive pixdims only and do not have an origin set. Part
of the reason for working this way is that the analyze standard does not
define how the origin/negative pixdims/orient fields etc interact and
also
the standard is not often met in practice. Hence we have found it
better
to only work with the simple case (zero origin, positive pixdims, no
orient)
for the most part and then more or less hard-code the avg152 coordinate
system in for the standard images we use.
At the present it is imperative that people check at their site which
way
around their images get converted with respect to left and right, as
this is
often impossible to tell by just looking at the images. Using phantom
images, or images of a person with a marker on one side is the best
method to check whether they are displayed the same way for left/right
as the standard image by fslview or slices. If this is the case then
the
default procedures can be used. If not, a left/right flip must be
performed
using avwswapdim prior to any other processing. Note that flirt will
never
flip left and right when finding a registration.
I don't know if this answers all of your questions. This issue is very
muddy
as analyze does not deal with it clearly. There is an extension of the
analyze
format hopefully coming soon which will help enormously in this regard,
but it isn't quite ready yet.
All the best,
Mark
On Friday, September 12, 2003, at 08:59 pm, Yasunari Tosa wrote:
> Hi, Flirt people:
>
> I found the answer to Question 1 and 2 by creating images
> with pixdim[1] > 0 and pixdim[1] < 0, but kept the data same
> I used flsview to see the images. I also used the LR marked
> image.
>
> Flirt expects the image to be -1 0 0
> 0 1 0
> 0 0 1
>
> It does not matter whether pixdim[1] <0 or >0.
> Flirt does not check.
>
> It is good to see that Flirt can handle 90 degree rotation.
>
> Thank you.
>
> Tosa
>
> On Thu, 2003-09-11 at 17:01, Yasunari Tosa wrote:
>> HI,
>>
>> I just downloaded "flirt Version 5.0". I noticed that
>> all the images in /etc/standard have pixdim[1] < 0, but
>> pixdim[2] and pixdim[3] are positive.
>> (Analyze format is so confusing with direction.
>> Sometimes, hist.orient is used. You have hist.orient = 0.)
>>
>> I really appreciate it very much if you could clarify how
>> you are using the pixdim[i] to denote the direction cosines.
>>
>> Question 1:
>> ===========
>> How are you using this information? Does this mean that
>> the direction cosine is of the following form?
>> x_r = -1 y_r = 0 z_r = 0
>> x_a = 0 y_a = 1 z_a = 0
>> x_s = 0 y_s = 0 z_s = 1
>> where I denoted RAS(right-anterior-posterior) direction
>> of the x, y, z axes component.
>>
>> What happens if pixdim[1] > 0? Is this the way?
>> x_r = 1 y_r = 0 z_r = 0
>> x_a = 0 y_a = 1 z_a = 0
>> x_s = 0 y_s = 0 z_s = 1
>> Or you never use this coordinate system.
>>
>> Question 2:
>> ===========
>> Can "flirt" handle the opposite case where e.g pixdim[1] > 0
>> where the direction cosine is given by
>> x_r = 1 y_r = 0 z_r = 0
>> x_a = 0 y_a = 1 z_a = 0
>> x_s = 0 y_s = 0 z_s = 1
>> (this is the average_305.mnc case where MNI defines
>> the orientation stored in the format) ?
>>
>> Or Do I have to convert the image so that I will have
>> pixdim[1] < 0?
>>
>> Depending on these, left and right are flipped.
>>
>> Question 3:
>> ===========
>> Can "flirt" handle the general cases where the direction cosine
>> of the src is, e.g.in the coronal direction but the reference is
>> in the horizontal direction (LAS or RAS) ?
>>
>> Thank you in advance for your reply
>>
>> Tosa
>> ====================================================
>> Yasunari Tosa, Ph.D.
>> R&D Enginner email: [log in to unmask]
>> Mass. General Hospital NMR Center phone:
>> Building 149
>> 13th Street fax :
>> Charlestown, MA 02129
> --
> Yasunari Tosa, Ph.D.
> R&D Enginner email: [log in to unmask]
> Mass. General Hospital NMR Center phone:
> Building 149
> 13th Street fax :
> Charlestown, MA 02129
|