Python Sparse data Analysis Package external MRI plugin.

This is the base reconstructor class for reconstruction. This class holds some parameters that are common for all MR Image reconstructors.

Parameters

fourier_op : object of class FFT, NonCartesianFFT or Stacked3DNFFT in

mri.operators

Defines the fourier operator F.

linear_op : object

Defines the linear sparsifying operator W. This must operate on x and have 2 functions, op(x) and adj_op(coeff) which implements the operator and adjoint operator. For wavelets, this can be object of class WaveletN or WaveletUD2 from mri.operators

regularizer_op : operator, (optional default None)

Defines the regularization operator for the regularization function H. If None, the regularization chosen is Identity and the optimization turns to gradient descent.

gradient_formulation : str between ‘analysis’ or ‘synthesis’,

default ‘synthesis’ defines the formulation of the image model which defines the gradient.

grad_class : Gradient class from mri.operators.

Points to the gradient class based on the MR Image model and gradient_formulation.

init_gradient_op : bool, default True

This parameter controls whether the gradient operator must be initialized right now. If set to false, the user needs to call initialize_gradient_op to initialize the gradient at right time before reconstruction

verbose : int, optional default 0

Verbosity levels

1 => Print basic debug information 5 => Print all initialization information 20 => Calculate cost at the end of each iteration. 30 => Print the debug information of operators if defined by class NOTE - High verbosity (>20) levels are computationally intensive.

extra_grad_args : Extra Keyword arguments for gradient initialization

This holds the initialization parameters used for gradient initialization which is obtained from ‘grad_class’. Please refer to mri.operators.gradient.base for reference. In case of sythesis formulation, the ‘linear_op’ is also passed as an extra arg

Notes

For the Analysis case, finds the solution for x of: ..math:: (1/2) * ||F x - y||^2_2 + mu * H (W x)

For the Synthesis case, finds the solution of: ..math:: (1/2) * ||F Wt alpha - y||^2_2 + mu * H(alpha)

initialize_gradient_op(**extra_args)[source]¶

reconstruct(kspace_data, optimization_alg='pogm', x_init=None, num_iterations=100, cost_op_kwargs=None, **kwargs)[source]¶

This method calculates operator transform.

Parameters

kspace_data : np.ndarray

the acquired value in the Fourier domain. this is y in above equation.

optimization_alg : str (optional, default ‘pogm’)

Type of optimization algorithm to use, ‘pogm’ | ‘fista’ | ‘condatvu’

x_init : np.ndarray (optional, default None)

input initial guess image for reconstruction. If None, the initialization will be zero

num_iterations : int (optional, default 100)

number of iterations of algorithm

cost_op_kwargs : dict (optional, default None)

specifies the extra keyword arguments for cost operations. please refer to modopt.opt.cost.costObj for details.

kwargs : extra keyword arguments for modopt algorithm

Please refer to corresponding ModOpt algorithm class for details. https://github.com/CEA-COSMIC/ModOpt/blob/master/ modopt/opt/algorithms.py