Add quick utils functions to get B and C matrices from off-resonance maps

[chaithyagr]

Currently, in

mri-nufft/src/mrinufft/operators/off_resonnance.py

Lines 19 to 58 in f5a6ba0

	class MRIFourierCorrected(FourierOperatorBase):
	"""Fourier Operator with B0 Inhomogeneities compensation.
	This is a wrapper around the Fourier Operator to compensate for the
	B0 inhomogeneities in the k-space.
	Parameters
	----------
	fourier_op: object of class FourierBase
	the fourier operator to wrap
	B: numpy.ndarray
	C: numpy.ndarray
	indices: numpy.ndarray
	backend: str, default 'cpu'
	the backend to use for computations. Either 'cpu' or 'gpu'.
	"""
	def __init__(self, fourier_op, B, C, indices, backend="cpu"):
	if backend == "gpu" and not CUPY_AVAILABLE:
	raise RuntimeError("Cupy is required for gpu computations.")
	if backend == "gpu":
	self.xp = cp
	elif backend == "cpu":
	self.xp = np
	else:
	raise ValueError("Unsupported backend.")
	self._fourier_op = fourier_op
	if not fourier_op.uses_sense:
	raise ValueError("please use smaps.")
	self.n_samples = fourier_op.n_samples
	self.n_coils = fourier_op.n_coils
	self.shape = fourier_op.shape
	self.smaps = fourier_op.smaps
	self.n_interpolators = len(C)
	self.B = self.xp.array(B)
	self.B = self.xp.tile(self.B, (self._fourier_op.n_samples // len(B), 1))
	self.C = self.xp.array(C)
	self.indices = indices

we just generate the fourier_op which takes the B and C values into account.

• It might help to also include a function in mrinufft.operators.off_resonance called get_interpolators_from_fieldmap.
• Also, it might further help to be able to get fourier correction possible directly from the init function itself like MRIFourierCorrected(fourier_op, b0_field, mask)
• It could be useful to allow for autodiff to pass through this new fourier op, currently we only support cupy or numpy, it might help to check if we have autograd enabled on based fourier_op and enable it as well (we just need to have tensors as torch tensors.
• A bit distant, but it could help to have the off-resonance estimation from k-space data as function. Although, this is a bit distant and I really feel it should belong to pysap-mri.

[paquiteau]

I fully agree with you.

Getting the off resonance map from the k space data could be done in the extras module, just as we propose it for the smaps estimation.

We may also wish to have autodiff support around this as well ?

[chaithyagr]

Sure. But note that the estimation isnt trivial, it is computationally expensive (for now). @Daval-G you think these codes can be open sourced though? I wanted to confirm that.

[philouc]

I agree with @paquiteau : Estimation of off-resonance effects from k-space data should be carried out separately, as it depends on the phase information usually extracted from the image domain and TE value.