This software segments structural magnetic resonance images automatically into anatomical regions using a database of segmented images (atlases) as a knowledge base.
MAPER exemplifies ensemble machine learning to approximate solutions to an ill-posed problem: there is no objective arbiter for drawing a boundary between anatomical regions in the brain on an in vivo image. MAPER achieves high consistency and accuracy with respect to manual reference segmentations.
Robustness is achieved by calculating an initial, coarse transformation between image-derived tissue probability maps, which is used as a starting point for registering the intensity images. Process yields are ca. 99.5% or higher (for example when segmenting ADNI baseline T1-weighted images using the Hammersmith Atlas Database). Segmentation results tend to be plausible even in severe brain atrophy and other abnormal brain configurations.
The rationale and principle are described in detail in the following paper.
Heckemann, R. A., Keihaninejad, S., Aljabar, P., Rueckert, D., Hajnal, J. V., Hammers, A., May 2010. Improving intersubject image registration using tissue-class information benefits robustness and accuracy of multi-atlas based anatomical segmentation. NeuroImage 51 (1), 221-227. http://dx.doi.org/10.1016/j.neuroimage.2010.01.072
If you use this software in your own work, please acknowledge MAPER by citing the above.
MAPER is based on earlier work on multi-atlas based segmentation:
Heckemann, R. A., Hajnal, J. V., Aljabar, P., Rueckert, D., Hammers, A., October 2006. Automatic anatomical brain MRI segmentation combining label propagation and decision fusion. NeuroImage 33 (1), 115-126. http://dx.doi.org/10.1016/j.neuroimage.2006.05.061
Since the 2010 paper, MAPER has been rewritten three times and ported to MIRTK for the registration steps. The principal idea remains the same, however.
Tested on Linux (NixOS 19.03, Ubuntu 16.04, CentOS 7) and on macOS (Big Sur -- needs Bash updated to version 5.1.4 or higher). Works well with multi-core and large-scale cluster architectures, as registering multiple atlas images to a target image is embarrassingly parallel.
For non-niche dependencies, cf. default.nix.
Clone or download & unpack, then test with
cd maper && export PATH=$PWD:$PATH
mkdir ~/testrun && cd ~/testrun
run-maper-example-generate.sh
# Modify run-maper-example.sh if and as desired
bash run-maper-example.sh
This downloads a mini-set of atlases with seven members and runs MAPER with one of the atlas images as the target.
Use the following to invoke MAPER for a single image using the mini-atlas
from the above example. The image is assumed to be a T1-weighted 3D
skullstripped MR, ie. every non-brain voxel is set to zero
intensity, and the image file is stored in ~/testrun/mybrain-T1w.nii.gz:
mkdir MAPER-MyBrain
printf "id, mri\nMyBrain, mybrain-T1w.nii.gz\n" >target.csv
launchlist-gen -src-description mini-atlas-n7r95/source-description.csv \
-tgt-description target.csv \
-output-dir MAPER-MyBrain
bash launchlist.sh
To parallelize the above onto seven threads, replace the last line with
cut -d ' ' -f 2- launchlist.sh | xargs -L 1 -P 7 maper
Use with the Hammersmith Atlas Database)
Download and unpack the atlas database in ~/atlas, then run
mkdir ~/atlas/ancillaries
hammers_mith-ancillaries.sh ~/atlas ~/atlas/ancillaries
This will download and unpack the ancillary data needed for MAPER in the
given location, including the source description csv file. Point
launchlist-gen to this file via the -src-description option.
In addition to the parallelization approach with xargs noted under
Instructions above, MAPER supports threaded execution of MIRTK
commands, if MIRTK is built with TBB support. This is less
memory-intensive than shell-level parallelization. Use the -threads
option to launchlist-gen and maper.
Feedback welcome at metrimorphics@soundray.de