bartorch.tools.wave

Contents

bartorch.tools.wave#

bartorch.tools.wave(maps, wave, kspace, *, r=None, b=None, i=None, s=None, c=None, t=None, e=None, g=False, f=False, H=False, v=False, w=False, l=False, **extra)#

Perform a wave-caipi reconstruction. Conventions: * (sx, sy, sz) - Spatial dimensions. * wx - Extended FOV in READ_DIM due to wave’s voxel spreading. * (nc, md) - Number of channels and ESPIRiT’s extended-SENSE model operator dimensions (or # of maps). Expected dimensions: * maps - ( sx, sy, sz, nc, md) * wave - ( wx, sy, sz, 1, 1) * kspace - ( wx, sy, sz, nc, 1) * output - ( sx, sy, sz, 1, md)

Runs bart wave.

Parameters:

  • maps (torch.Tensor) – Input array.

  • wave (torch.Tensor) – Input array.

  • kspace (torch.Tensor) – Input array.

  • r (float | None) – Soft threshold lambda for wavelet or locally low rank. (-r)

  • b (int | None) – Block size for locally low rank. (-b)

  • i (int | None) – Maximum number of iterations. (-i)

  • s (float | None) – Step size for iterative method. (-s)

  • c (float | None) – Continuation value for IST/FISTA. (-c)

  • t (float | None) – Tolerance convergence condition for iterative method. (-t)

  • e (float | None) – Maximum eigenvalue of normal operator, if known. (-e)

  • g (bool) – use GPU (-g)

  • f (bool) – Reconstruct using FISTA instead of IST. (-f)

  • H (bool) – Use hogwild in IST/FISTA. (-H)

  • v (bool) – Split result to real and imaginary components. (-v)

  • w (bool) – Use wavelet. (-w)

  • l (bool) – Use locally low rank across the real and imaginary components. (-l)

  • **extra (Any) – Further BART flags, passed through by name.

Returns:

output

Return type:

torch.Tensor

Contents