pygrog: A PyTorch-based Package for Implicit GROG Interpolation

pygrog is a PyTorch library for implementing implicit GRAPPA operator gridding (GROG) with field modeling. This package enables efficient interpolation of non-Cartesian MRI data onto Cartesian grids using deep learning. With pygrog, you can set up and train neural networks that replace conventional interpolation with a multi-source model, improving the quality of MRI reconstructions.

Features

• Calibration: Extract low-resolution k-space data or synthesize calibration data.
• Trajectory Pre-processing: Generate and upsample displacement vectors from non-Cartesian trajectories.
• GRAPPA Training: Set up and train neural networks for GROG interpolation.
• Interpolation: Apply trained models to new non-Cartesian datasets for interpolation.
• Sparse-to-Dense Gridding: Convert interpolated sparse data to dense Cartesian grids.

Installation

To install the package, clone the repository and run:

git clone https://github.com/your-username/pygrog.git
cd pygrog
pip install -e .

Requirements

• Python >= 3.7
• PyTorch >= 1.8
• numpy
• finufft / cufinufft / torchkbnufft / sigpy (for NUFFT operations)

Quick Start

Here’s a simple example showing how to use pygrog:

import torch
from pygrog.calibration.calibration_data import CalibrationData

# Step 1: Calibration
kspace_data = torch.rand(64, 64)
trajectory = torch.rand(64, 2)  # Example 2D non-Cartesian trajectory
calibration = CalibrationData(kspace_data, trajectory)
low_res_data = calibration.extract_low_res((16, 16))

print("Low-resolution k-space data shape:", low_res_data.shape)

Documentation

Documentation for pygrog follows the Diátaxis framework:

• Tutorials: Step-by-step guides for getting started.
• How-to Guides: Instructions for specific use cases.
• Explanations: Background on implicit GROG and interpolation concepts.
• References: API reference for all classes and functions.

To build the documentation locally, navigate to the docs/ directory and run:

make html

Testing

Tests for pygrog are located in the tests/ directory and use pytest. To run the tests, execute:

pytest tests/

Project Structure

The project follows a src layout:

pygrog/
├── src/
│   └── pygrog/
│       ├── calibration/
│       ├── trajectory_preprocessing/
│       ├── grappa_training/
│       ├── interpolation/
│       └── grid_reconstruction/
├── tests/
├── examples/
└── docs/

Contributing

We welcome contributions! Please read our CONTRIBUTING.rst for guidelines on how to contribute to pygrog.

License

pygrog is licensed under the MIT License. See the LICENSE file for more details.

Contact

For questions or support, please open an issue on the project's GitHub page or contact the authors.