Feature comparison

What each toolbox provides itself, checked against its source and documentation in October 2026 (and not since: newer releases may have added what a cell says is missing) (BART 731bfd3, SigPy 0.1.27, Gadgetron 1d14c4c, MRpro 6830010, current MRIReco.jl and MIRT.jl). "via X" means the feature comes from another package.

Feature	BART	SigPy	Gadgetron	MRIReco.jl	MIRT.jl	MRpro	Ristretto
Iterative SENSE	✅	✅	✅ (GPU)	✅	✅ (operator + solver)	✅	✅
g-factor / noise propagation	❌	❌	✅ (GRAPPA g-map, pseudo-replica)	❌	❌	❌	✅ (pseudo-replica)
GRAPPA	❌	❌	✅	❌	❌	❌	✅
SPIRiT	❌	❌	✅ (incl. ℓ₁-SPIRiT)	❌	❌	❌	✅
ESPIRiT	✅	✅	via BART	✅ (Cartesian)	❌	❌ (Walsh, Inati)	✅
JSENSE / NLINV	✅ NLINV	✅ JSENSE	via BART	❌	❌	❌	❌
Calibrationless structured low-rank	✅ SAKE	❌	❌	❌	❌	❌	✅ SAKE, LORAKS
Noise prewhitening	✅	✅	✅	✅	❌	✅	✅
Coil compression	✅ SVD, GCC, ROVir	❌	✅ PCA	✅ SVD, GCC	✅ PCA	✅ PCA	✅ SVD, GCC
Non-Cartesian (NUFFT)	✅	✅	✅	✅	✅	✅	✅
Density compensation	❌ (given weights)	✅ Pipe–Menon	✅	✅	radial only	✅ Voronoi	✅ Pipe–Menon, Voronoi
B₀ off-resonance correction	✅	✅ time-segmented	✅ spiral	✅ time-segmented	building blocks	✅	❌
Partial Fourier	homodyne	❌	✅ homodyne, POCS	❌	❌	❌	✅ homodyne, POCS, phase-constrained
Gradient delay estimation	✅ RING	❌	❌	❌	❌	❌	✅ RING, opposing spokes
Total variation	✅	✅	✅	✅	edge-preserving	✅	✅
Wavelets	✅	✅	✅	✅	✅	✅	✅
Low-rank, locally low-rank	✅ (+ multi-scale)	❌	❌	✅ (via RegularizedLeastSquares)	❌	❌	✅ (+ multi-scale)
Low-rank + sparse decomposition	matrix completion	❌	❌	❌	example script	❌	✅ any regularizers
TGV	✅ (TGV, ICTGV)	❌	❌	❌	❌	❌	✅
Subspace / temporal basis	✅	❌	❌	✅	❌	✅	✅
Plug-and-play priors	✅	❌	❌	✅ (via RegularizedLeastSquares)	❌	❌	✅
Unrolled networks	✅ MoDL, VarNet	❌	❌	❌	❌	❌	❌
Simultaneous multi-slice	model-based	❌	❌	❌	❌	❌	❌
Quantitative parameter mapping	✅ model-based	❌	✅ cardiac T₁, T₂	❌	❌	✅	❌
GPU	✅ CUDA	✅ CuPy	✅ CUDA	✅ GPUArrays	❌	✅ PyTorch	✅ GPUArrays (tested on CUDA)
Scanner streaming (MRD)	streaming import	❌	✅ server	file I/O	❌	file I/O	MRD read and write
Image export	PNG, DICOM	❌	DICOM, MRD	NIfTI (via ImageUtils)	AVS `.fld`	NIfTI, DICOM	NIfTI, DICOM, MRD

PROPELLER/BLADE reconstruction is in none of them. What Ristretto lacks is on its roadmap.