[Insight-developers] Update on GE5 Image Reading (STILL NEED SCANS!)

[kent williams]

Mea Maxima Culpa -- the GE file formats Image I/O stuff is the very first
ITK code I wrote, and there is more than a little ugliness involved there
unfortunately.  I was converting code from BRAINS2 to the ITK way.

Anyway, I have a fix that I can check in that makes images look much better.
As near as I can figure, the sorting of slice files was backwards in some
cases.  It was even worse than that really -- itk::GE5ImageIO was looking at
fields in the GE5/Signa/Genesis file header that don't really mean what the
code thought it meant.

Someone (Bill Lorensen is one suspect) added the code to generate direction
cosines directly from the GE5 file header, so I started using that instead
of undocumented fields in the header to recover the orientation.  That
DirCosine code is as far as I know correct.

But, the test images were coming in swapped in one dimension or another.
What I finally figured out is that the code that sorted the 2D files before
moving data into a 3D buffer wasn't using the right sort order, at all. So I
added code to set the sort order right, based on the image orientation.

I THINK I have it licked but I'm still not finding good, unambiguous test
data.  I can tell where the top of the skull and the nose are, but bilateral
symmetry means I can't tell when there's a L/R swap.

Fucang was kind enough to find me a few good sample files.  But does anyone
have any GE5 files where there's some indication of which is left and right?
Something like the 'vitaman E cap on the left temple' trick would be ideal.

Speaking of sample files:

This directory http://128.196.98.224/Public/Imaging/Event/MGH/New/ is
supposed to be an example of an axially oriented GE5 file.  If you load it
into Slicer3 (for example) it looks way way ugly. Not clear to me why, but
Slicer3 displays the volume as a skewed frustrum, as though every coronal
slice is offset in the Y direction from it's neighbors.