API reference

Exact contracts of the public objects: parameters, defaults, shapes, units, conventions and restrictions. Concepts are in Explanations, complete workflows in the examples.

Page	Module	Contents
Array functions and settings	bartorch	Fourier and non-uniform Fourier transforms, density compensation, k-space windows, wavelet transforms, thresholding, array functions, runtime settings
BART commands as functions	bartorch.tools	BART commands as functions: simulation, sampling, calibration, preprocessing, registration, metrics, low-rank completion; corrections and rigid motion estimation
Linear operators	bartorch.linop	Linear operators, operator algebra and MRI encoding operators
Nonlinear operators	bartorch.nlop	Nonlinear operators, MRI and signal models, Gauss-Newton methods
Optimization	bartorch.optim	Iterative solvers, iteration blocks and data scaling
Regularization and denoising	bartorch.priors	Regularization terms, plug-and-play priors and BART’s denoisers
Reconstruction pipelines	bartorch.apps	BART reconstruction pipelines assembled from operators and solvers
Learning	bartorch.learning	Networks for complex images, unrolled iterations, patchwise execution, self-supervised splitting, uncertainty and training stages
Interoperability	bartorch.interop	DeepInverse physics adapter
File I/O	bartorch.io	CFL files, ISMRMRD raw data, DICOM and NIfTI images
Command line	bartorch.cli	The bartorch command line

BART correspondence

bartorch accepts BART’s command lines unchanged: bartorch <command> ... takes the arguments bart <command> ... takes (Command line). In Python, arrays are C-order tensors whose shape is BART’s dimension vector reversed, coils first, (coils, z, y, x) (Data layout), and axes are indices, never bitmasks.

BART	bartorch
bart <command> ...	bartorch <command> ..., or bartorch.tools.<command>(...)
bart fft -u -i 7	bartorch.ifft(x, (-3, -2, -1), unitary=True)
bart nufft	bartorch.nufft, linop.NUFFT
bart pics -R W:...	apps.pics(..., regularizers=priors.Wavelet(...))
bart pics -R T:...	apps.pics(..., regularizers=priors.TotalVariation(...))
bart pics -t traj	apps.pics(..., traj=traj)
bart moba, bart mobafit	apps.moba, apps.mobafit
bart ecalib, bart nlinv, bart traj, bart phantom	bartorch.tools.ecalib, .nlinv, .traj, .phantom

BART’s fft is unnormalised unless -u; linop.FFT is unitary.