MRIMixin#

class deepinv.physics.MRIMixin[source]#

Bases: object

Mixin base class for MRI functionality.

Base class that provides helper functions for FFT and mask checking.

check_mask(mask=None, three_d=False, device='cpu', **kwargs)[source]#

Updates MRI mask and verifies mask shape to be B,C,…,H,W where C=2.

Parameters:

mask (torch.nn.parameter.Parameter, torch.Tensor) – MRI subsampling mask.
three_d (bool) – If False the mask should be min 4 dimensions (B, C, H, W) for 2D data, otherwise if True the mask should have 5 dimensions (B, C, D, H, W) for 3D data.
device (torch.device, str) – mask intended device.

crop(x, crop=True)[source]#

Center crop 2D image according to img_size.

This matches the RSS reconstructions of the original raw data in deepinv.datasets.FastMRISliceDataset.

If img_size has odd height, then adjust by one pixel to match FastMRI data.

Parameters:

x (torch.Tensor) – input tensor of shape (…,H,W)
crop (bool) – whether to perform crop, defaults to True

static fft(x, dim=(-2, -1), norm='ortho')[source]#

Centered, orthogonal fft

Parameters:

x (torch.Tensor) – input image of complex dtype of shape [B,…] where … is all dims to be transformed
dim (tuple) – fft transform dims, defaults to (-2, -1)
norm (str) – fft norm, see docs for torch.fft.fftn(), defaults to “ortho”

static from_torch_complex(x)[source]#

[B,…,H,W] complex -> [B,2,…,H,W] real

static ifft(x, dim=(-2, -1), norm='ortho')[source]#

Centered, orthogonal ifft

Parameters:

x (torch.Tensor) – input kspace of complex dtype of shape [B,…] where … is all dims to be transformed
dim (tuple) – fft transform dims, defaults to (-2, -1)
norm (str) – fft norm, see docs for torch.fft.fftn(), defaults to “ortho”

im_to_kspace(x, three_d=False)[source]#

Convenience method that wraps fft.

Parameters:

x (torch.Tensor) – input image of shape (B,2,…) of real dtype
three_d (bool) – whether MRI data is 3D or not, defaults to False

Returns:

Tensor: output measurements of shape (B,2,…) of real dtype

Return type:

Tensor

kspace_to_im(y, three_d=False)[source]#

Convenience method that wraps inverse fft.

Parameters:

y (torch.Tensor) – input measurements of shape (B,2,…) of real dtype
three_d (bool) – whether MRI data is 3D or not, defaults to False

Returns:

Tensor: output image of shape (B,2,…) of real dtype

Return type:

Tensor

static rss(x, multicoil=True, three_d=False)[source]#

Perform root-sum-square reconstruction on multicoil data, defined as

\[\operatorname{RSS}(x) = \sqrt{\sum_{n=1}^N |x_n|^2}\]

where \(x_n\) are the coil images of \(x\), \(|\cdot|\) denotes the magnitude and \(N\) is the number of coils. Note that the sum is performed voxel-wise.

Parameters:

x (torch.Tensor) – input image of shape (B,2,…) where 2 represents real and imaginary channels
multicoil (bool) – if True, assume x is of shape (B,2,N,…), and reduce over coil dimension N too.

static to_torch_complex(x)[source]#

[B,2,…,H,W] real -> [B,…,H,W] complex