BART commands as functions

bartorch.tools exposes BART’s command-line applications that have no operator or pipeline counterpart here, one function per command, run in this process. Array arguments and results are C-order tensors in BART’s dimension order reversed (Data layout and conventions), and axis arguments are axis indices rather than BART bitmasks; the results of a BART command are not recorded by autograd. Functions marked wrapped have a hand-written signature; the others are generated from BART’s declaration of the command and take its options under their long names. Reconstructions are bartorch.apps, or an encoding from bartorch.linop under a solver from bartorch.optim; Interfaces and execution compares the two routes. The last two sections, corrections and rigid motion, hold functions and classes with no BART command behind them.

Simulation

Object	Description
phantom	Analytical phantom as an image or as k-space, optionally with coils or along a trajectory (wrapped)
coils	Analytical coil sensitivities
fakeksp	K-space from an image and sensitivities
noise	Additive complex Gaussian noise

Sampling and trajectories

Object	Description
traj	Cartesian, radial and golden-angle trajectories in grid units (wrapped)
grid	Sampling grid coordinates in image space or k-space
pattern	Sampling pattern of a k-space array
poisson	Poisson-disc sampling pattern
upat	Regular undersampling pattern with a fully sampled centre
raga	Indices of a RAGA (rational approximation of the golden angle) radial ordering
grog	Radial data gridded onto a Cartesian grid by GROG
psf	Point spread function of a trajectory
wavepsf	Wave-CAIPI point spread function in hybrid space
estdims	Image dimensions implied by a non-Cartesian trajectory
estdelay	Gradient-delay estimation from radial data
trajcor	Gradient-delay correction of a trajectory
rmfreq	Removal of an angle-dependent frequency from radial data
bin	Binning of data by labels, including respiratory and cardiac quadrature binning
ssa	Singular spectrum analysis (SSA-FARY) of a time series

Coil calibration and compression

Object	Description
ecalib	Sensitivities by ESPIRiT (wrapped)
caldir	Sensitivities directly from the low-resolution k-space centre (wrapped)
calmat	Calibration matrix of the k-space centre
ecaltwo	Second stage of ESPIRiT calibration
walsh	Walsh coil combination, for use with ecaltwo
nlinv	Image and sensitivities jointly by nonlinear inversion (wrapped)
ncalib	Sensitivities from non-Cartesian data by nonlinear inversion (ENLIVE)
cc	Coil compression matrix (SVD, geometric or ESPIRiT)
ccapply	Application of a coil compression matrix
rovir	Coil compression by region-optimized virtual coils (ROVir)
whiten	Noise prewhitening from a noise measurement, optionally returning the whitening matrix and the noise covariance (wrapped)
estvar	Noise variance of white Gaussian noise
phasepole	Detection of phase poles in sensitivities

Preprocessing, registration and metrics

Object	Description
fovshift	Field-of-view shift of k-space by a linear phase (wrapped)
homodyne	Homodyne partial-Fourier reconstruction of asymmetrically sampled k-space
affine_transform	Resampling on an affinely mapped grid (wrapped)
warp	Resampling through a displacement field (wrapped)
register_affine	Affine registration by mutual information (wrapped)
register_nonrigid	Non-rigid registration by greedy SyN or TV-L1 optical flow (wrapped)
estimate_shift	Sub-voxel translation from the phase of the cross-spectrum (wrapped)
nrmse	Normalized root-mean-square error, optionally after least-squares scaling (wrapped)
mse	Mean squared error (wrapped)
psnr	Peak signal-to-noise ratio of the magnitudes (wrapped)
ssim	Structural similarity of the magnitudes (wrapped)
roi_stat	Statistic over a region of interest (wrapped)

Low-rank completion

Object	Description
sake	SAKE: low-rank matrix completion of k-space
lrmatrix	Multi-scale low-rank matrix completion

Corrections

Corrections applied to data or images outside the reconstruction. Bias field and gradient nonlinearity correction resample with SimpleITK (pip install 'bartorch[correct]'), and spiral deblurring uses Triton on a CUDA device when it is installed. Susceptibility correction runs PyHySCO, which is GPL-3.0-only, is not distributed with bartorch and is imported only when correct_susceptibility is called (pip install 'bartorch[pyhysco]').

Object	Description
epi_ramp_operator	Band-limited resampling of a ramp-sampled EPI readout onto a uniform grid
estimate_epi_phase	Odd/even phase of an EPI readout, fitted to a three-line navigator
correct_lines	Reversal of EPI lines into forward readout order, with the odd/even phase removed
bias_field_correct	Receive-field (bias) correction by N4
GradientCoefficients	Spherical-harmonic coefficient table of a gradient coil
CoefficientAccessor	Protocol for a coefficient table that is not a file
Gradunwarp	Gradient nonlinearity correction of images from a coefficient table
field_map_from_phase	Off-resonance field map from the phase of single-echo coil images
ReadoutTiming	Readout time of a spiral arm as a function of k-space radius
SpiralTransfer	Low-rank factorization of the off-resonance transfer of a spiral readout
fit_transfer	Fit of a SpiralTransfer to a readout’s time map
deblur	Off-resonance deblurring of a spiral image
correct_susceptibility	Susceptibility distortion correction from a reversed phase-encoding pair, by PyHySCO
SusceptibilityCorrection	Corrected pair and displacement field from correct_susceptibility

Rigid motion from navigators

Plane reconstruction by the non-uniform transform of this package, rigid registration by SimpleITK (pip install 'bartorch[motion]'), and a constant-velocity extended Kalman filter over the registered poses.

Object	Description
reconstruct_navigator	Magnitude images of a navigator’s planes by density-compensated adjoint NUFFT
RigidRegistration	Multi-resolution rigid registration of magnitude images
RigidMotionEstimate	2D or 3D rigid transform with its covariance
RigidMotionEKF	Constant-velocity extended Kalman filter over registered rigid poses
NavigatorMotionTracker	Six-degree-of-freedom pose from a navigator’s 2D planes, filtered over time