PyMyelinPSO is a Python framework for particle swarm optimization (PSO)–based inversion of in-vivo and atlas MRI data. It enables voxel-wise parameter estimation in single and joint inversion modes, currently targeting T2 and T2* relaxation modeling.
PyMyelinPSO implements a regularization-free global search inversion strategy for multi-echo MRI data. Instead of relying on explicit regularization constraints, the framework performs stochastic multi-cycle optimization, enabling:
- Bias-free myelin water fraction (MWF) quantification
- Voxel-wise uncertainty assessment derived from inversion ill-posedness
- Pareto-based solution exploration
- Systematic evaluation of preprocessing effects (e.g., denoising, Gibbs correction) on quantitative myelin metrics
This approach supports clinically interpretable parameter maps and transparent sensitivity analysis of modeling assumptions.
-
Voxel-wise inversion of T2 and T2* signals (real and complex-valued)
-
Single and joint inversion workflows
-
Slice-parallel full-volume processing
-
Pixel-wise iterative mode for detailed Pareto front analysis
-
Iteration-test mode for convergence diagnostics
-
Multi-core parallelization is implemented via Python’s ProcessPoolExecutor combined with memory-mapped arrays for computational efficiency and scalability.
-
Automated generation of:
- synthetic decay curves
- model vector arrays
- full-volume parameter maps (e.g., MWF, misfit)
- Pareto plots
- convergence diagnostics
The current version supports T2 and T2* (real and complex-valued) inversion. Full T1 inversion support is planned for future releases.
For complete instructions on how to use PyMyelinPSO, refer to the Example Notebook, which demonstrates the computation and visualization of solutions from different available runnig modes for single- and joint inversion PSO.
This project is licensed under the GNU Affero General Public License v3.0 - see the LICENSE.md file for details.
If PyMyelinPSO is advancing your reasearch, please cite as:
Kobe, M., Dega, S., Stöcker, T., & Paasche, H. (2026). PyMyelinPSO: Particle Swarm Optimization of in-vivo MRI data for unbiased quantitative Myelin Mapping (1.0). Zenodo. https://doi.org/10.5281/zenodo.18709789
This research was funded by the Helmholtz Association, Initiative and Networking Fund, Helmholtz Imaging Projects (HIP), funding code: ZT-I-PF-4-077 (Helmholtz Imaging Project “JIMM2”).