A fast implementation of the GRAPPA algorithm for parallel MRI reconstruction, built on PyTorch.
This project leverages GPU acceleration and efficient tensor operations to significantly reduce reconstruction time, making GRAPPA practical as a preprocessing step in deep learning pipelines.
- GPU Acceleration: PyTorch backend enables massive speedups compared to NumPy-based implementations.
- Matrix-Optimized: Vectorized operations replace Python loops for efficiency.
- Fast Pattern Detection: Unique undersampling patterns identified with optimized tensor operations.
- Flexible Integration: Easily embedded into PyTorch training workflows (e.g., fastMRI, MONAI).
Typical speedups: 30–40× faster than CPU-based PyGRAPPA.
To build and install this package locally in editable mode, follow these steps:
pip install build
python -m build
pip intall -e .| Method | Hardware | Avg Runtime (s) | Speedup |
|---|---|---|---|
| PyGRAPPA (CPU) | i7-8550U | 7.57 | 1× |
| Proposed (GPU) | MX150 | 0.18 | 42× |
Below is an example reconstruction on a T2-weighted brain MRI slice:
Experimental setup: kernel size = 5×5, number of coils = 20, acceleration factor R = 4, λ = 0.1.
Results:
- PSNR = 35.09 dB
- SSIM = 0.885
The reconstructed image closely matches the ground truth, with only minor residuals visible in the difference map.
We welcome contributions to improve this implementation further. If you have suggestions, bug reports, or feature requests, please open an issue or submit a pull request.
This work builds upon the pygrappa repository by Nicholas McKibben.