Machine Learning Enabled Brain Segmentation for Small Animal Image Registration
These are the content files used to generate scientific communication materials for the project originally titled “Improving Registration in Small Animal Brain Imaging”.
Dependencies
A full list of unambiguously identified dependency constraints is specified here (following the Package Manager Specification format). On a Gentoo Linux system, dependencies can be automatically managed with the included install script:
su -
cd /path/to/mlebe
cd .gentoo
./install.sh
Manual Data Download (only if automated Gentoo dependency management is unavailable)
While other dependencies will very likely be available from your distribution's own package manager, the data package of this publication is probably not. You can manually install it via the following commands:
wget https://zenodo.org/record/3601531/files/irsabi_bidsdata-1.4.tar.xz
tar xf irsabi_bidsdata-1.4.tar.xz
mv irsabi_bidsdata-1.4 /usr/share/irsabi_bidsdata
The latter command may require superuser access. Total install time will take upwards of an hour on personal computers with no prior neuroimaging software deployment.
Pretrained classifiers can be downloaded here or via:
wget https://zenodo.org/record/4031286/files/3D_unet_EPI.zip
wget https://zenodo.org/record/4031286/files/3D_unet_RARE.zip
unzip 3D_unet_EPI.zip
unzip 3D_unet_RARE.zip
Compilation Instructions
This is a RepSeP -style document. The data processing step is run asynchronously from the document compilation, and you may choose to reproduce either the top-level statistics (“demo” reproduction) or the entire analysis starting from the raw data (“full analysis stack” reproduction).
The data processing for the full analysis stack will by default take place in ~/.scratch, which we suggest can be created as a symlink to a volume which has more space (at least 400GB):
mkdir /mnt/largevolume/.scratch
ln -s /mnt/largevolume/.scratch ~/.scratch
Top-Level Analysis
cd /path/to/mlebe
./compile.sh
This analysis may take up to 5 minutes on personal computers.
Full Analysis Stack
cd /path/to/mlebe
./produce.sh
This analysis may take up over 24 hours on personal computers.
When running the workflow, make sure to specify the path to the folder containing the classifiers in one of the config files from prepare/configs. The path to the workflow config can be specified in specified prepare/make_config.py.