Segmentation of hippocampal subfields
Author: Koen Van Leemput
See also: GEMS
Automated segmentation of the subfields of the hippocampus, using models built from manual segmentations of the right hippocampus in 0.38x0.38x0.8 mm^3 in vivo MRI scans in 10 subjects. For the left hippocampus, the model for the right side is simply mirrored. The exact algorithm is described in the paper:
Automated Segmentation of Hippocampal Subfields from Ultra-High Resolution In Vivo MRI. K. Van Leemput, A. Bakkour, T. Benner, G. Wiggins, L.L. Wald, J. Augustinack, B.C. Dickerson, P. Golland, B. Fischl. Hippocampus, vol. 19, no. 6, pp. 549-557, June 2009
To generate the hippocampal subfield segmentations as part of the normal FreeSurfer processing stream, simply add the "-hippo-subfields" flag to the recon-all script:
recon-all -s bert -all -hippo-subfields
It is also possible to only run the hippocampal subfield segmentation for a subject that has already undergone the standard volumetric FreeSurfer pipeline (in particular, the program needs the FreeSurfer-generated files "aseg.mgz", "nu.mgz" and "talairach.xfm"):
recon-all -s bert -hippo-subfields
The output consists of a collection of images $SUBJECTS_DIR/bert/mri/posterior_* that indicate each voxel's posterior probability of belonging to different subregions. The estimated volumes of these subregions can be obtained using the "kvlQuantifyHippocampalSubfieldSegmentations.sh" script. Doing
will create the text files "nonPartialVolumeStatsLeft.txt" and "nonPartialVolumeStatsRight.txt" in the directory $SUBJECTS_DIR, containing tables that summarize the volumes of the hippocampal subfield segmentations of the left and right side, respectively, in all of the subjects in $SUBJECTS_DIR. The reported volumes are given in number of voxels, where each voxel is of the size "0.5x0.5x0.5mm^3". It is expected that the "-hippo-subfields" flag has been used for each subject in $SUBJECTS_DIR; if this is not the case the "kvlQuantifyHippocampalSubfieldSegmentations.sh" script will fail and no (or incomplete) output files will be generated.
The estimated hippocampal subfield volumes can also be computed for one specific subject only: For the right hippocampus, the command is
cd $SUBJECTS_DIR/bert/mri kvlQuantifyPosteriorProbabilityImages $FREESURFER_HOME/data/GEMS/compressionLookupTable.txt \ posterior_right_* posterior_Right-Hippocampus.mgz
and for the left
cd $SUBJECTS_DIR/bert/mri kvlQuantifyPosteriorProbabilityImages $FREESURFER_HOME/data/GEMS/compressionLookupTable_left.txt \ posterior_left_* posterior_Left-Hippocampus.mgz
Visualizing the subregion segmentations on top of the MRI data can be achieved by doing
cd $SUBJECTS_DIR/bert/mri freeview nu.mgz \ -p-labels posterior_left_* posterior_Left-Hippocampus.mgz \ -p-labels posterior_right_* posterior_Right-Hippocampus.mgz \ -p-prefix posterior_ -p-lut $FREESURFER_HOME/FreeSurferColorLUT.txt
The segmentation results are given as probability maps: a (unsigned 8-bit) value of 0 means a probability is 0, 225 means a probability of 1, and everthing in between scales linearly.
Since the console output generated during subfields segmentation is long, the log is written in an alternate file - $SUBJECTS_DIR/bert/scripts/hippo-subfields.log. The file recon-all.log simply shows the invocation of various commands during the subfields generation.
Report bugs to <email@example.com>