GPU-accelerated python implementation of ODF filtering algorithm for asymmetric ODF estimation in diffusion MRI. The method is presented in Poirier et al. (2024).
Important
AODF filtering is now available as a Scilpy script (scil_sh_to_aodf.py)! In addition to the CPU and GPU OpenCL implementations, Scilpy also implements a pure-python version of the filtering. We recommend new and current aodf-toolkit users to migrate to Scilpy where it will be more easily maintainable.
Note
Filter calibration is not yet available through Scilpy but should be integrated in the near future.
For installing aodf-toolkit, see Installation section below.
If you use the method, please cite. A BibTeX entry is given below.
@article{poirier_unified_2024,
title = {A unified filtering method for estimating asymmetric orientation distribution functions},
journal = {NeuroImage},
pages = {120516},
year = {2024},
issn = {1053-8119},
doi = {https://doi.org/10.1016/j.neuroimage.2024.120516},
url = {https://www.sciencedirect.com/science/article/pii/S1053811924000119},
author = {Charles Poirier and Maxime Descoteaux},
}
The application can be installed using pip. The application is tested with python 3.10. We recommend installing the application inside a virtual environment (see python venv instructions here). To install, run the following command from the project root:
pip install -e .
Once installed, the application can be called from the command line:
aodf_filter in_sh.nii.gz out_sh.nii.gz
For tuning the hyper-parameter values (filter sigmas) to the data, an automatic calibration script is provided:
aodf_calibrate_filter in_sh.nii.gz calibrate_output.csv
The script outputs a csv file containing the self-supervised loss associated with different hyper-parameter values.
For a description of available options, use --help option.