def forward( self, im ):
im = np.fft.fftshift(im,self.axes)
ksp = np.fft.fft2(im,s=None,axes=self.axes)
ksp = np.fft.ifftshift(ksp,self.axes)
#try to match the dims of ksp and mask
if len(ksp.shape) is not len(self.mask.shape):
#try to match the dimensions of ksp and mask
ksp_out_shape, mask_out_shape = dim_match(ksp.shape, self.mask.shape)
mksp = np.multiply(ksp.reshape(ksp_out_shape),\
self.mask.reshape(mask_out_shape))#apply mask
else:
mksp = np.multiply(ksp,self.mask)#apply mask
return mksp
def allcoil_nufft(ktraj, dcf, kdata, ncoils, im_shape):
im = np.zeros((im_shape[0], im_shape[1], im_shape[2], ncoils),
dtype=kdata.dtype)
kdatashape, dcfshape = ut.dim_match(kdata.shape, dcf.shape)
kdata = np.multiply(kdata.reshape(kdatashape),
dcf.reshape(dcfshape)).squeeze()
c = kdata.reshape((np.prod(kdata.shape[0:2]), ncoils)).squeeze()
ktraj = np.pi * ktraj / ktraj.flatten().max() #2.0 *
x = ktraj[0, :].flatten()
y = ktraj[1, :].flatten()
z = ktraj[2, :].flatten() #*(1.0*im_shape[0]/im_shape[2])
#im = nft.nufft3d1_gaussker(x, y, z, c, im_shape[0], im_shape[1], im_shape[2], df=1.0, eps=1E-5)
im = nft_cuda.nufft3d1_gaussker_cuda(x,
y,
z,
c,
im_shape[0],
im_shape[1],
im_shape[2],
df=1.0,
eps=1E-5,
gridfast=0)
return im
def density_weighting( self, kdata, dcf ):
kdatashape, dcfshape = dim_match(kdata.shape, dcf.shape)
kdata = np.multiply(kdata.reshape(kdatashape), dcf.reshape(dcfshape)).squeeze()
return kdata
def forward( self, im_sos ):
sens_out_shape, im_out_shape = dim_match(self.sens.shape,im_sos.shape)
#appying sensitivity profile is sens*im
return np.multiply(im_sos.reshape(im_out_shape),\
self.sens.reshape(sens_out_shape))
def backward( self, im_coils ):
sens_out_shape, im_out_shape = dim_match(self.sens.shape,im_coils.shape)
# coil combination is sum(conj(sens)*im)
return np.sum(np.multiply(im_coils.reshape(im_out_shape),\
np.conj(self.sens).reshape(sens_out_shape))\
, axis=self.coil_axis, keepdims = True)
