This is the official implementation of the deepbet paper.

deepbet is a neural network based tool, which achieves state-of-the-art results for brain extraction of T1w MR images of healthy adults, while taking ~1 second per image.

Usage

After installation, there are three ways to use deepbet

1. deepbet-gui runs the Graphical User Interface (GUI)

2. deepbet-cli runs the Command Line Interface (CLI)

deepbet-cli -i /path/to/inputs -o /path/to/output/brains

3. Run deepbet directly in Python

from deepbet import run_bet

input_paths = ['path/to/sub_1/t1.nii.gz', 'path/to/sub_2/t1.nii.gz']
brain_paths = ['path/to/sub_1/brain.nii.gz', 'path/to/sub_2/brain.nii.gz']
mask_paths = ['path/to/sub_1/mask.nii.gz', 'path/to/sub_2/mask.nii.gz']
tiv_paths = ['path/to/sub_1/tiv.csv', 'path/to/sub_2/tiv.csv']
run_bet(input_paths, brain_paths, mask_paths, tiv_paths, threshold=.5, n_dilate=0, no_gpu=False)

Besides the input paths and the output paths

• brain_paths: Destination filepaths of input nifti files with brain extraction applied
• mask_paths: Destination filepaths of brain mask nifti files
• tiv_paths: Destination filepaths of .csv-files containing the total intracranial volume (TIV) in cm³
  • Simpler than it sounds: TIV = Voxel volume * Number of 1-Voxels in brain mask

you can additionally do

• Fine adjustments via threshold: deepbet internally predicts values between 0 and 1 for each voxel and then includes each voxel which is above 0.5. You can change this threshold (e.g. to 0.1 to include more voxels).
• Coarse adjustments via n_dilate: Enlarges/shrinks mask by successively adding/removing voxels adjacent to mask surface.

and choose if you want to use GPU (only NVIDIA supported) for speedup

• no_gpu: deepbet automatically uses the NVIDIA GPU if available. If you do not want that, set no_gpu=True.

Installation

For accelerated processing via GPU, it is recommended to first install PyTorch separately via a command customized for your system.

Then the package itself can be installed via

pip install deepbet

Due to this issue, the GUI can look ugly, which can be resolved via

conda install -c conda-forge tk=*=xft_*

Citation

If you find this code useful in your research, please consider citing

@article{deepbet,
    title = {deepbet: Fast brain extraction of T1-weighted MRI using Convolutional Neural Networks},
    journal = {Computers in Biology and Medicine},
    volume = {179},
    pages = {108845},
    year = {2024},
    issn = {0010-4825},
    doi = {https://doi.org/10.1016/j.compbiomed.2024.108845},
    url = {https://www.sciencedirect.com/science/article/pii/S0010482524009302},
    author = {Lukas Fisch and Stefan Zumdick and Carlotta Barkhau and Daniel Emden and Jan Ernsting and Ramona Leenings and Kelvin Sarink and Nils R. Winter and Benjamin Risse and Udo Dannlowski and Tim Hahn},
}