January 2006 FreeSurfer Release

Goals

• Post new 'stable' release, containing reliable auto-segmentation functionality, and bug fixes
• Make available open-source Freesurfer code (minus Numerical Recipes for C code)
• Basic automated testing of nightly-build 'dev' code base

Work Items

1. aseg (xiao)
   1. make aseg work for data originating from different pulse sequences or manufactuers than the training data used to build the atlas. currently, the subcortical atlas is built from Siemens scans, and the aseg results will be quite poor if the input data is from a GE scanner. we have sw that can improve this situation a lot, but it's not very stable, and this cross-sequence option will make the aseg on Siemens data worse.
   2. testing to optimize the default parameters for the subcortical segmentation.
   3. testing of a new subcortical atlas (asegs should be better than the current atlas).

      Item #	% complete	Expected date of completion
      1	100%	done
      2	?	?
      3	?	?

   B. recon-all improvements (doug)
   1. linear pipeline: in recon-all, re-naming of stuff so that files don't get overwritten later in the stream.
   2. test accuracy of aseg and aparc w/ mri_pretess.

      Item #	% complete	Expected date of completion
      1 coding	90	12/15/05
      2 debugging	40	12/16/05
      3 testing	0	1/3/05

   C. miscellaneous (doug)
   1. total intercranial volume, using randy's method to align to talairach with skull.
   2. fixing tkmedit so that it implements -overlay-reg-identity properly.
   3. mri_surf2surf - mapping across hemispheres.
   4. mri_glmfit (new version of mris_glm) - mostly just need to document.
   5. mris_preproc - surface-based preprocessor script for mri_glmfit (does not exist but not hard).
   6. correction for multiple comparisons (mri_mcsim and mri_surfcluster).
   7. improve editing of wm with aseg

      Item #	% complete	Expected date of completion
      1	100	done
      2	?	?
      3	?	?
      4	?	?
      5	?	?
      6	?	?
      7	100	done

   D. talairach failure detection, incorporated into recon-all (nick, laurence, doug)
   1. create Matlab-compiled binary and test in recon-all stream.
   2. code in C and test in recon-all stream.

      Item #	% complete	Expected date of completion
      1	?	?
      2	?	?

   E. automated testing of nightly build (kevin, nick)
   1. basic mean/variance tests of recon'd bert data
   2. mean/variance tests of buckner 40
   3. aseg tests (xiao)
   4. verifying talairach coordinates, when tal xform is known good, of: (nick, david)
      1. individual surfaces. b. individual volumes. c. average surfaces. d. average volumes.

      Item #	% complete	Expected date of completion
      1	?	?
      2	?	?
      3	?	?
      4	?	?

   F. fix manual labelling (ev)

Work Item Notes

E.4: verifying talairach coordinates The problem being addressed in this work is that the talairach coordinates supplied by tkmedit are based on the alignment to the MNI305 brain, which is poorly aligned to the Talairach brain (see http://www.nil.wustl.edu/labs/kevin/man/answers/mnispace.html). The root of the problem is that the Talairach brain was never MRI scanned, and furthermore was labelled post-mordem, so it is very difficult to map a scanned brain to true Talairach coordinates. Yet we would like some assurance that the tal coordinates that Freesurfer supplies (in tkmedit and elsewhere) are reasonably accurate.