View source for 2008-08-15-email-atlas-generation

......
Thinking a little bit more about this, I decided to open up the discussion.

I had a similar request from somebody else during the meeting that I attended this week. He is part of the group that works with Chris Wyatt, but works on human, not monkey. I believe that we will see  many such requests, since EM is one of the key distinguishers for our "stuff". That is why we should work on these final pieces and the corresponding documentation.

Raphael has about 10 subjects where he has manually labeled components of the orbits on the left and on the right for each subject. He would like to use that as a training set to build an atlas that can be used to drive the EM segmenter.

We have the rest of EM packaged nicely for sophisticated end-users, but there is no guideline for end-users on how to build the shape atlases to drive the EM segmenter. The data needs to be parked in something like xnat so that we have a chance to develop a standard mechanism to do this.

We should have a discussion during an upcoming engineering tcon on that topic.

Ron
.....
I had to do some dusting before I found the answer:

1) The script I generated the PCA atlases with is located at
/projects/birn/pohl/slicer2/Modules/vtkEMPrivateSegment/projects/miccai05/AtlasGeneration/GeneratePCA-Atlas-SpecialAtlas4

The script does the following:
1.)  Affine alignment (9 parameters) of all cases to template case and apply transformation to segmentations
2.) Generate Sign Distance Map for each segmentation
3.) Run PCA on Sign Distance maps

Here are two comments:
- I believe to preserve the properties of the PCA atlas we have to rigidly align the atlas to a new case.
- Alternatively we could non-rigidly align the new case to the template, such as Diffeomorphic Demons, segment the aligned case and than map the results back to the original coordinate system of the new case

I hope this helps

Kilian