• AtlasSubjects

This page is for NMR Center use. It lists the location of the subcortical and cortical atlas subject training and test sets, and instructions on how to create (train) and test these atlases.

Subcortical Segmentation - Aseg Atlas

The 'official' subcortical atlases distributed with Freesurfer are named 'RB_all_YYYY-MM-DD.gca' and 'RB_all_withskull_YYYY-MM-DD.gca', where YYYY-MM-DD is the year/month/day of when the atlas was built. In the 'recon-all' script, the atlases are named near the beginning of the file, like this:

# For subcortical segmentation
set GCA      = RB_all_2007-08-08.gca
set GCASkull = RB_all_withskull_2007-08-08.gca

If new atlases are created (instructions are next), then the .gca files need to be added to the Freesurfer CVS repository to the dev/distribution/average directory, the Makefile.am file updated, and recon-all updated with the new names.

See also Jenni's notes on the manual edits made to the training subjects.

Aseg Create

The 39 subjects used to create the two .gca files are found here:

cd /autofs/space/dijon_032/users/salat/RBnew
source scripts/subjects.csh
echo $SUBJECTS_DIR
echo $SUBJECTS
echo $#SUBJECTS

The /scripts/subjects.csh file will set the SUBJECTS env var to the list of the 41 subjects composing the atlas. Each subject has a file called 'seg_edited.mgz' in its mri directory, which is a symbolic link to the most recent valid edited aseg volume. Demographic info for those 39 subjects is found in the file 'demographics'.

To build the subcortical atlases, which should be performed on the cluster machine 'seychelles' because so many processes are necessary:

ssh seychelles
  cd /autofs/space/dijon_032/users/salat/RBnew
  source scripts/subjects.csh
  source /usr/local/freesurfer/nmr-dev-env
  rebuild_gca_atlas.csh &

Then wait about two days for it to complete. The location of the log file is displayed after rebuild_gca_atlas.csh finishes submitting all the jobs. That log file should be monitored for the location of the output files and their status. Note that 'dev' is used, whereas 'stable' might be preferable, depending on whether the atlas is intended for testing or public release (in stable).

Aseg Test

See AsegTestNotes

Cortical Parcellation - Aparc Atlas

The 'official' cortical atlas distributed with Freesurfer is named in the recon-all script as follows:

# For cortical registration, as found in $AvgCurvTifPath/$hemi.$AvgCurvTif
set AvgCurvTifPath = "${FREESURFER_HOME}/average"
set AvgCurvTif = average.curvature.filled.buckner40.tif

# For cortical parcellation atlas, as found in:
# $FREESURFER_HOME/average/$hemi.$GCS
set GCS = curvature.buckner40.filled.desikan_killiany.YYYY-MM-DD.gcs

If new atlases are created (instructions are next), then the .gcs and .tif files need to be added to the Freesurfer CVS repository to the dev/distribution/average directory, the Makefile.am file updated, make_average_surface and recon-all updated with the new names.

Aparc Create

The 40 training subjects are here:

cd /autofs/space/amaebi_026/users/buckner_cortical_atlas
source scripts/subjects.csh
setenv SUBJECTS_DIR $PWD
echo $SUBJECTS_DIR
echo $SUBJECTS
echo $#SUBJECTS

Each subject has two files called label/?h.aparc_edited.annot which are the manually labeled cortical surface files.

To create new atlases:

cd /autofs/space/amaebi_026/users/buckner_cortical_atlas
setenv SUBJECTS_DIR $PWD
build_desikan_killiany_gcs.csh rh
build_desikan_killiany_gcs.csh lh

The build_desikan_killiany_gcs.csh script runs mris_ca_train. Upon completion, the average directory will contain the new .gcs files (for each hemi). A log file is kept in the average directory. New .gcs atlas files, after testing, should be put in the source tree at dev/distribution/average, and the recon-all should be updated to use it (see GCS var).

Aparc Test

Training subjects jackknife testing

See parc_atlas_jackknife_test --help

Test subjects classification testing

80 manually labeled (and reconed) test subjects are found here:

cd /autofs/space/sake_033/users/nicks/aparc_test
source scripts/subjects.csh
echo $SUBJECTS_DIR
echo $SUBJECTS
echo $#SUBJECTS

In this set, these are the areas that are labeled:

bankssts
entorhinal
fusiform
inferiorparietal
inferiortemporal
isthmuscingulate
middletemporal
parahippocampal
posteriorcingulate
precuneus
rostralanteriorcingulate
superiorparietal
superiortemporal
supramarginal
temporalpole
transversetemporal

To test, use mris_compute_parc_overlap to compare each subjects manual parcellation (their ?h.aparc_edited.annot file) to their automatically generated parcellation (?h.aparc.annot):

foreach s ($SUBJECTS)

    echo $s

    mris_compute_parc_overlap --s $s --hemi rh \
    --annot1 aparc --annot2 aparc_edited \
    --label-list valid_labels.txt \
    >& $s/label/mris_compute_parc_overlap.rh.log &

    mris_compute_parc_overlap --s $s --hemi lh \
    --annot1 aparc --annot2 aparc_edited \
    --label-list valid_labels.txt \
    >& $s/label/mris_compute_parc_overlap.lh.log

end

Note the argument --label-list valid_labels.txt, where the file valid_labels.txt is a text file listing the aforementioned areas that were labeled. mris_compute_parc_overlap only looks at those labels in computing Dice and the mean min distance measures.

Once the preceding loop is complete, then some overall stats can be extracted. To do so, run:

cd $SUBJECTS_DIR
./scripts/get_stats

and this will extract the mean and std overall Dice values for the left and right hemis, and ditto for the mean-min distance values.

To view those results in bar-chart form, comparing against prior test runs, edit the file scripts/aparc_test_run_plots.m, adding the new data pairs where instructed. Then open matlab and run:

cd scripts
matlab &
matlab>> aparc_test_run_plots

Verify that the current test run results are not too deviant from prior test runs.

Demographics

The master Buckner demographics excel spreadsheet is found here: /autofs/space/dijon_028/users/jpacheco/new_stream_buckner/master_buckner_demographics_092906.xls