fastmri-tools 0.1.0

Standalone fastMRI prostate T2 and diffusion regridding, GRAPPA filling, and complex IFFT outputs.

fastmri_tools

Standalone reconstruction helper for fastMRI prostate T2 and diffusion HDF5 files.

The project is intentionally self-contained: it does not import anything from the parent repository. Runtime dependencies are ordinary pip packages listed in pyproject.toml.

What it does

For each input MRI HDF5 file, the pipeline:

1. Loads kspace, calibration_data, and the ISMRMRD header.
2. Detects T2 or diffusion data, or uses the sequence passed on the command line.
3. Applies trapezoidal EPI regridding for diffusion data when header parameters are present. T2 data keeps its Cartesian readout grid.
4. Fills missing samples with GRAPPA using the calibration data.
5. Computes a centered complex inverse FFT from the filled k-space.
6. Writes complex outputs with the same dimensions as the original input k-space: (averages, slices, coils, readout, phase).

The output HDF5 contains:

• kspace_regridded: complex k-space after diffusion regridding, or a copy of the input k-space for T2.
• kspace_grappa: complex GRAPPA-filled k-space.
• image_complex: complex centered inverse FFT of kspace_grappa.
• image_complex_coil_combined: diffusion only, when coil_sens_maps is available in the source file. This is still complex and has shape (averages, slices, readout, phase).

Install

From PyPI, once the first release has been published:

python -m pip install fastmri_tools

From GitHub:

python -m pip install git+https://github.com/meis-01/fastmri_tools.git

For local development:

python -m venv .venv
.\.venv\Scripts\python -m pip install -U pip
.\.venv\Scripts\python -m pip install -e ".[dev]"

Run one file

fastmri-tools reconstruct `
  --input C:\path\to\file.h5 `
  --output C:\path\to\recon_complex.h5 `
  --sequence auto

Use --sequence t2 or --sequence dwi to override auto-detection.

Run a folder

fastmri-tools reconstruct `
  --input C:\path\to\h5_folder `
  --output C:\path\to\output_folder `
  --sequence auto

The command writes one output HDF5 per input file.

Quick synthetic check

This creates a tiny synthetic T2-like file and reconstructs it:

fastmri-tools make-synthetic --output .\synthetic_t2.h5 --sequence t2
fastmri-tools reconstruct --input .\synthetic_t2.h5 --output .\synthetic_t2_recon.h5 --sequence t2

Inspect the output:

fastmri-tools inspect .\synthetic_t2_recon.h5

The module entry points also work:

python -m fastmri_tools inspect .\synthetic_t2_recon.h5

Publish a release

This repository includes GitHub Actions for CI and PyPI publishing.

1. Create a PyPI project or pending publisher for fastmri_tools.
2. Configure PyPI Trusted Publishing with:
   • Owner: meis-01
   • Repository name: fastmri_tools
   • Workflow name: publish.yml
   • Environment name: pypi
3. Create and publish a GitHub release such as v0.1.0.

GitHub Actions will build the source distribution and wheel, then publish them to PyPI without storing a PyPI API token in GitHub secrets.

Notes

• The complex inverse FFT is intentionally not converted to RSS magnitude. This preserves phase and coil information.
• Output spatial dimensions match the loaded k-space dimensions. No T2 zero padding or center cropping is applied.
• Diffusion regridding is skipped automatically if the ISMRMRD header does not contain the needed EPI timing fields.