Dear Tommy,
> However, so far we are having trouble getting good results out of
> it. We
> have tested pointflirt by creating matrix files using the voxels
> reported by
> FSLview, and also by using the coordinates (in mm) as reported by
> FSLview.
> We have even attempted to input scaled coordinates by subtracting the
> coordinate values when the voxel values are at (0,0,0.) One obvious
> anomaly
> is that the fourth column of the output matrix has a value that
> measures in
> the 1000's.
Could you please send us an example of the coordinate files that you
put in and the resulting output matrix?
> 1) Would you mind explaining scaled coordinates and their
> application as it
> relates to this?
Scaled coordinates is simply coordinates in mm with the centre of the
first voxel as (0,0,0). From looking at the code it looks like it
should apply to this. If there is a qform/sform set in either of your
images the mm coordinates that are reported by fslview will (as far as
I can tell) not be directly applicable for pointflirt. The easiest
then might be to just use the voxel coordinates.
> 2) Might it have to do with plugging in both positive and negative
> values
> when it can only take positive?
It _can_ take both negative and positive values, but in practice the
valid range for both voxel coordinates and scaled coordinates is
positive values, so a negative value would indicate that one or both
of your images has an sform/qform, in which case I recommend using
voxel coordinates instead.
> 3) Are four points necessary to use pointflirt? Or can we just plug
> in 3
> points, which is all we need to align in 3D space?
It should be possible to use just 3 points, but in that case you need
to make sure you get your rows and columns right (since they are now
in principle interchangeable). The format should be
x1 x2 ...
y1 y2 ...
z1 z2 ...
etc.
Good Luck Jesper
|