deepmr.recon.recon_lstsq

Contents

deepmr.recon.recon_lstsq#

deepmr.recon.recon_lstsq(data, head, mask=None, niter=1, prior=None, prior_ths=0.01, prior_params=None, solver_params=None, lamda=0.0, stepsize=1.0, basis=None, nsets=1, device=None, cal_data=None, toeplitz=True, use_dcf=True)[source]#

Classical MR reconstruction.

Parameters:
  • data (np.ndarray | torch.Tensor) – Input k-space data of shape (nslices, ncoils, ncontrasts, nviews, nsamples).

  • head (deepmr.Header) – DeepMR acquisition header, containing traj, shape and dcf.

  • mask (np.ndarray | torch.Tensor, optional) – Sampling mask for Cartesian imaging. Expected shape is (ncontrasts, nviews, nsamples). The default is None.

  • niter (int, optional) – Number of recon iterations. If single iteration, perform simple zero-filled recon. The default is 1.

  • prior (str | deepinv.optim.Prior, optional) –

    Prior for image regularization. If string, it must be one of the following:

    • "L1Wav": L1 Wavelet regularization.

    • "TV": Total Variation regularization.

    The default is None (no regularizer).

  • prior_ths (float, optional) – Threshold for denoising in regularizer. The default is 0.01.

  • prior_params (dict, optional) – Parameters for Prior initializations. See deepmr.prox(). The defaul it None (use each regularizer default parameters).

  • solver_params (dict, optional) – Parameters for Solver initializations. See deepmr.optim(). The defaul it None (use each solver default parameters).

  • lamda (float, optional) – Regularization strength. If 0.0, do not apply regularization. The default is 0.0.

  • stepsize (float, optional) – Iterations step size. If not provided, estimate from Encoding operator maximum eigenvalue. The default is None.

  • basis (np.ndarray | torch.Tensor, optional) – Low rank subspace basis of shape (ncontrasts, ncoeffs). The default is None.

  • nsets (int, optional) – Number of coil sensitivity sets of maps. The default is ``1.

  • device (str, optional) – Computational device. The default is None (same as data).

  • cal_data (np.ndarray | torch.Tensor, optional) – Calibration dataset for coil sensitivity estimation. The default is None (use center region of data).

  • toeplitz (bool, optional) – Use Toeplitz approach for normal equation. The default is True.

  • use_dcf (bool, optional) – Use dcf to accelerate convergence. The default is True.

Returns:

Reconstructed image of shape:

  • 2D Cartesian: ``(nslices, ncontrasts, ny, nx).

  • 2D Non Cartesian: ``(nslices, ncontrasts, ny, nx).

  • 2D Non Cartesian: ``(nslices, ncontrasts, ny, nx).

  • 3D Non Cartesian: ``(ncontrasts, nz, ny, nx).

Return type:

img np.ndarray | torch.Tensor