deepmr.linops.FFTGramOp#
- class deepmr.linops.FFTGramOp(*args: Any, **kwargs: Any)[source]#
Self-adjoint Sparse Fast Fourier Transform operator.
K-space sampling mask, if provided, is expected to to have the following dimensions:
2D MRI:
(ncontrasts, ny, nx)
3D MRI:
(ncontrasts, nz, ny)
Input and output images are expected to have the following dimensions:
2D MRI:
(nslices, nsets, ncoils, ncontrasts, ny, nx)
3D MRI:
(nsets, ncoils, ncontrasts, nz, ny)
where
nsets
represents multiple sets of coil sensitivity estimation for soft-SENSE implementations (e.g., ESPIRIT), equal to1
for conventional SENSE andncoils
represents the number of receiver channels in the coil array.Methods
__init__
([mask, basis, device])Initiate the linear operator.
forward
(x)Apply Toeplitz convolution (
SparseFFT.H * SparseFFT
).to
(*args, **kwargs)Copy to different devices
Attributes
H
Return adjoint linear operator.