Data module

Transform function template

activemri.data.transform_template(kspace: List[numpy.ndarray] = None, mask: torch.Tensor = None, ground_truth: torch.Tensor = None, attrs: List[Dict[str, Any]] = None, fname: List[str] = None, slice_id: List[int] = None)

Template for transform functions.

Parameters

  • kspace (-) – A list of complex numpy arrays, one per k-space in the batch. The length is the batch_size, and array shapes are H x W x 2 for single coil data, and C x H x W x 2 for multicoil data, where H denotes k-space height, W denotes k-space width, and C is the number of coils. Note that the width can differ between batch elements, if num_cols is set to a tuple when creating the environment.

  • mask (-) – A tensor of binary column masks, where 1s indicate that the corresponding k-space column should be selected. The shape is batch_size x 1 x maxW, for single coil data, and batch_size x 1 x 1 x maxW for multicoil data. Here maxW is the maximum k-space width returned by the environment.

  • ground_truth (-) – A tensor of ground truth 2D images. The shape is batch_size x 320 x 320.

  • attrs (-) – A list of dictionaries with the attributes read from the fastMRI for each image.

  • fname (-) – A list of the filenames where the images where read from.

  • slice_id (-) – A list with the slice ids in the files where each image was read from.

Returns

A tuple with any number of inputs required by the reconstructor model.

Return type

tuple(Any…)

Custom transforms

activemri.data.transforms.py

Transform functions to process fastMRI data for reconstruction models.

activemri.data.transforms.fastmri_unet_transform_multicoil(kspace=None, mask=None, ground_truth=None, attrs=None, fname=None, slice_id=None)

Transform to use as input to fastMRI’s Unet model for multicoil data.

This is an adapted version of the code found in fastMRI.

activemri.data.transforms.fastmri_unet_transform_singlecoil(kspace=None, mask=None, ground_truth=None, attrs=None, fname=None, slice_id=None)

Transform to use as input to fastMRI’s Unet model for singlecoil data.

This is an adapted version of the code found in fastMRI.

activemri.data.transforms.raw_transform_miccai2020(kspace=None, mask=None, **_kwargs)

Transform to produce input for reconstructor used in Pineda et al. MICCAI’20.

Produces a zero-filled reconstruction and a mask that serve as a input to models of type activemri.models.cvpr10_reconstructor.CVPR19Reconstructor. The mask is almost equal to the mask passed as argument, except that high-frequency padding columns are set to 1, and the mask is reshaped to be compatible with the reconstructor.

Parameters

  • kspace (np.ndarray) – The array containing the k-space data returned by the dataset.

  • mask (torch.Tensor) – The masks to apply to the k-space.

Returns

A tuple containing:

  • torch.Tensor: The zero-filled reconstructor that will be passed to the reconstructor.

  • torch.Tensor: The mask to use as input to the reconstructor.

Return type

tuple