NEMAIECPhantomDataset#

class deepinv.utils.phantoms.NEMAIECPhantomDataset(size=128, activities=None, background=0.25, lung=0.0, normalize=True, transform=None, use_dict_output=False)[source]#

Bases: Dataset

Dataset for the NEMA IEC body phantom.

Each item contains one channel. A torch.utils.data.DataLoader adds the batch dimension.

Note

See deepinv.utils.phantoms.generate_nema_iec_phantom() for phantom geometry and generation details Daube-Witherspoon et al.[1].

Parameters:

  • size (int or tuple[int, int, int]) – Size of a square transverse slice, or volume shape (D, H, W).

  • activities (list[float]) – List of 6 sphere activity levels, ordered by increasing diameter. Defaults to [1.0] * 6.

  • background (float) – activity level of the uniform background.

  • lung (float) – activity level of the central cold insert.

  • normalize (bool) – if True, rescale so that the maximum value is 1.

  • transform (Callable) – Transformation to apply to the output image.

  • use_dict_output (bool) – Whether to return output as dict with key “x” or a bare Tensor (default: False).

References: