Freesurfer Subparcellation

Freesurfer annotations and labels describe swaths of cortical regions, defined anatomically during the recon-all process, or functionally or statistically (e.g., significant GLM clusters). Tools exist to further subdivide these regions to the desired level of granularity.

Using mris_divide_parcellation

This program divides one or more parcellations into divisions perpendicular to the long axis of the label. The number of divisions can be specified in one of two ways, depending upon the nature of the fourth argument.

First, a splitfile can be specified as the fourth argument. The splitfile is a text file with two columns. The first column is the name of the label in the source annotation, and the second column is the number of units that label should be divided into. The names of the labels depends upon the source parcellation. For aparc.annot and aparc.a2005.annot, the names can be found in $FREESURFER_HOME/FreeSurferColorLUT.txt. For aparc.annot, the labels are between the ranges of 1000-1034. For aparc.a2005s.annot, the labels are between the ranges of 1100-1181. The name for the label is the name of the segmentation without the 'ctx-lh'. Note that the name included in the splitfile does not indicate the hemisphere. For example, 1023 is 'ctx-lh-posteriorcingulate'. You should put 'posteriorcingulate' in the splitfile. Eg, to divide it into three segments, the following line should appear in the splitfile:

posteriorcingulate 3

Only labels that should be split need be specified in the splitfile.

The second method is to specify an area threshold (in mm^2) as the fourth argument, in which case each label is divided until each subdivision is smaller than the area threshold.

The output label name will be the original name with _divN appended, where N is the division number. N will go from 2 to the number of divisions. The first division has the same name as the original label.

Sample Split File (Functional Clusters)

In this example, I have carried out a one-sample group mean (OSGM) analysis and did a cluster-size threshold. This saved a number of clusters in two .annot files (one in each hemisphere). There are also some *.cluster.summary files in this directory. Reading these files, I see the smallest clusters are about 120mm². I would like to subdivide the larger clusters so that they are all between about 100-150mm². I load the corresponding .annot file in Freesurfer just to verify the cluster names (the summary file lists 10 clusters, but the names do not follow a consistent naming convention). From the file, I see that clusters 1,2,3,4,6 and 8 need to be subdivided. Simple division by the desired cluster size tells me the number of divisions I want for each cluster:

echo cluster-001 6 > rsplittable.txt
echo cluster-002 7 >> rsplittable.txt
echo cluster-003 5 >> rsplittable.txt
echo cluster-004 5 >> rsplittable.txt
echo cluster-006 2 >> rsplittable.txt
echo cluster-008 2 >> rsplittable.txt

Repeat the process for the lh clusters, creating an lsplittable.txt file.

The group analysis used the fsaverage surface. Since mris_divide_parcellation assumes we're subdividing a particular subject's annotation, it makes sense to copy the source .annot files into the fsaverage label/ folder (giving the appropriate ?h prefix)

cp cache.th30.abs.sig.ocn.annot \
$SUBJECTS_DIR\fsaverage\label\rh.cache.th30.abs.sig.ocn.annot

Then call the program thus (assuming that rsplittable.txt is in your current working directory):

mris_divide_parcellation fsaverage rh \ 
cache.th30.abs.sig.ocn.annot rsplittable.txt \
functional_subclusters

Program output:

reading colortable from annotation file...
colortable with 11 entries read (originally none)
interpreting 4th command line arg as split file name
dividing cluster-001 (1) into 6 parts
dividing cluster-002 (2) into 7 parts
dividing cluster-003 (3) into 5 parts
dividing cluster-004 (4) into 5 parts
dividing cluster-006 (6) into 2 parts
dividing cluster-008 (8) into 2 parts
allocating new colortable with 31 additional units...
saving annotation to functional_subclusters

Repeat for lh.

Note: the first time I ran this, I have no idea where the rh.functional_subclusters.annot file went! Running the lh and re-running the rh subparcellation generated the expected files in the fsaverage/label directory.

Check Subparcellation

Presumably the python wrapper does this iteratively because all subdivisions are done along the same axis. As I discovered, a large number of subdivisions may be realized as a series of thin bands (below). In this case, it might be best to subdivide larger regions over a series of calls, as I describe below.

Iterative Subparcellation

If banding among your subdivisions is a problem, here's an example of the iterative approach using region size, rather than a split file:

First, break down clusters into 400mm² chunks:

mris_divide_parcellation fsaverage lh \
cache.th30.abs.sig.ocn.annot 
400 400functional_subclusters

Next, break down the 400mm² chunks into sub-200mm² regions:

mris_divide_parcellation fsaverage lh \
400functional_subclusters.annot \
200 200functional_subclusters

Because each call to mris_divide_parcellation bisects the region along it's longest axis, successive function calls at each iteration produce increasingly less elongated subparcellations.

Transfer Subparcellation to Other Subjects

Once satisfied with the source subparcellation, the scheme can be mapped to other surfaces using mri_surf2surf (this needs to be done individually for the lh and rh surfaces -- below is an example command for the left):

mri_surf2surf \
--srcsubject fsaverage \
--trgsubject FS_T1_501 \
--hemi lh \
--sval-annot $SUBJECTS_DIR/fsaverage/label/lh.200functional_subclusters.annot \
--tval $SUBJECTS_DIR/FS_T1_501/label/lh.200functional_subclusters.annot

Pro Tip: Use a global variable (e.g. SUBN) and use $SUBN in place of your subject in trgsubject and tval. Better yet, use the clustmap.sh shell script.

You may want to pull out overlapping voxels for T1 and T2 data (like a venn diagram) before mapping the surfaces to each subject, here.