ESPIRiT

PyTorch-based ESPIRiT coil sensitivity calibration for MRI.

Single codebase that runs on CPU, CUDA GPU, and Apple Silicon (MPS) — no separate code paths needed.

Notice

This package contains a PyTorch translation of the ESPIRiT implementation from the BART (Berkeley Advanced Reconstruction Toolbox), © 2013–2026 The Regents of the University of California and BART Developer Team. BART is licensed under the BSD 3-Clause License. See https://codeberg.org/mrirecon/bart.

Usage

1) CLI

uvx espirit kspace.npy

2) Python

uv add espirit

import numpy as np
import torch
from espirit import espirit

# NumPy
kspace_np = np.load("kspace.npy")
csm_np = espirit(kspace_np)

# PyTorch 
kspace_pt = torch.randn(8, 24, 24, 24, dtype=torch.complex64)
csm_pt = espirit(kspace_pt)

Options

csm = espirit(
    kspace,             # (n_coils, *spatial_dims)
    calib_size=24,      # calibration region size
    kernel_size=6,      # sliding-window kernel size
    threshold=0.001,     # singular-value threshold
    mask_threshold=0.8, # eigenvalue mask threshold
    normalize=True,     # RSS=1 normalization
    rotphase=True,      # remove phase ambiguity
    device=None,        # cuda, mps, or cpu (auto-detect when None)
    output_device=None, # final CSM device; use "cpu" to reduce 3D GPU memory
    verbose_memory=False,
)

For large 3D datasets, output_device="cpu" moves completed sensitivity-map slices to CPU immediately. This reduces peak GPU memory while keeping the calibration and eigenmap calculations on the selected compute device. Tensor inputs return a tensor on output_device; NumPy inputs always return a NumPy array.

Device support

Device	Backend	Notes
`cpu`	NumPy/MKL	Always available
`cuda`	NVIDIA GPU	Requires CUDA toolkit
`mps`	Apple Metal	macOS with Apple Silicon

The same code runs on all devices — PyTorch handles dispatch automatically.

Development Status
- 3 - Alpha
Intended Audience
- Science/Research
Programming Language
- Python :: 3
- Python :: 3.12

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

espirit-0.2.0.tar.gz (19.5 kB view details)

Uploaded Aug 25, 2026 Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

espirit-0.2.0-py3-none-any.whl (12.5 kB view details)

Uploaded Aug 25, 2026 Python 3

File details

Details for the file espirit-0.2.0.tar.gz.

File metadata

Download URL: espirit-0.2.0.tar.gz
Upload date: Aug 25, 2026
Size: 19.5 kB
Tags: Source
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for espirit-0.2.0.tar.gz
Algorithm	Hash digest
SHA256	`caa70aebe74b23d4200e54fd53a9e86063782cb565636bd0f15d4082a8edaf3d`
MD5	`e7f3f7cc1baa4b50d683040c35b99cd1`
BLAKE2b-256	`15952ee47f0ed8d2d25efb646d8316b2417600d09de3bf45ef5d29d4ea1e05fa`

See more details on using hashes here.

Provenance

The following attestation bundles were made for espirit-0.2.0.tar.gz:

Publisher: python-publish.yml on oscarvanderheide/espirit

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Statement:
- Statement type: https://in-toto.io/Statement/v1
- Predicate type: https://docs.pypi.org/attestations/publish/v1
- Subject name: espirit-0.2.0.tar.gz
- Subject digest: caa70aebe74b23d4200e54fd53a9e86063782cb565636bd0f15d4082a8edaf3d
- Sigstore transparency entry: 2590498720
- Sigstore integration time: Aug 25, 2026
Source repository:
- Permalink: oscarvanderheide/espirit@f0fc3c89317c8bf3daaa643b0646312d617f5169
- Branch / Tag: refs/tags/v0.2.0
- Owner: https://github.com/oscarvanderheide
- Access: public
Publication detail:
- Token Issuer: https://token.actions.githubusercontent.com
- Runner Environment: github-hosted
- Publication workflow: python-publish.yml@f0fc3c89317c8bf3daaa643b0646312d617f5169
- Trigger Event: release