generate_pet_phantom#

deepinv.utils.phantoms.generate_pet_phantom(img_shape, mu_value=0.01, add_spheres=True, add_inner_cylinder=True, r0=0.45, r1=0.28, oversampling_factor=4, device='cpu')[source]#

Generate a 2D or 3D PET-like phantom and its corresponding attenuation map.

The phantom mimics a simplified PET scanner body phantom consisting of:

  • Outer elliptical body: a large elliptical cylinder (or ellipsoid in 3D) filled with uniform emission activity (value 1.0) and attenuation mu_value. Its semi-axes are r0 (along the first spatial dimension) and r1 (along the second spatial dimension), expressed as fractions of the respective image size.

  • Inner cold rod (optional, add_inner_cylinder=True): a small elliptical cylinder at the centre of the body with lower emission activity (value 0.25) and reduced attenuation (mu_value / 3), representing a low-activity insert.

  • Hot and cold spheres (optional, add_spheres=True): four pairs of spheres placed at two axial offsets inside the body. Hot spheres (value 2.5) simulate lesion-like regions with elevated metabolic activity; cold spheres (value 0.25) simulate photopenic defects (e.g. cysts or necrotic tissue).

The function generates the phantom at oversampling_factor times the requested resolution and then downsamples by averaging, reducing partial-volume effects at the boundaries.

For 2D inputs (img_shape = (H, W)), the phantom is generated as a 3D volume and the central axial slice is returned.

Parameters:

  • img_shape (tuple) – Shape of the output image (H, W) or volume (D, H, W).

  • mu_value (float) – Attenuation coefficient of the outer body (default: 0.01).

  • add_spheres (bool) – If True, hot and cold spheres are added inside the body (default: True).

  • add_inner_cylinder (bool) – If True, a low-activity inner rod is added at the centre of the body (default: True).

  • r0 (float) – Fractional semi-axis of the outer ellipse along the first spatial dimension (D or H for 3D/2D inputs). Must satisfy 0 < r0 <= 0.5 so that the ellipse fits within the image. Default: 0.45.

  • r1 (float) – Fractional semi-axis of the outer ellipse along the second spatial dimension (H or W for 3D/2D inputs). Must satisfy 0 < r1 <= 0.5 so that the ellipse fits within the image. If add_spheres=True, r1 should be at least ~0.25 to ensure the off-centre spheres remain inside the body. Default: 0.28.

  • oversampling_factor (int) – Upsampling factor used during phantom generation to reduce partial-volume artefacts (default: 4).

  • device (str or torch.device) – Device on which tensors are allocated (default: "cpu").

Returns:

Tuple (x_em, x_att) where both tensors have shape (1, 1, H, W) for 2D inputs or (1, 1, D, H, W) for 3D inputs. x_em is the emission activity map and x_att is the attenuation map.

Return type:

tuple[torch.Tensor, torch.Tensor]

Raises:

ValueError – If r0 or r1 are outside the valid range (0, 0.5].

Example:

>>> from deepinv.utils.phantoms import generate_pet_phantom
>>> x_em, x_att = generate_pet_phantom(img_shape=(64, 64, 32))
>>> print(x_em.shape, x_att.shape)
torch.Size([1, 1, 64, 64, 32]) torch.Size([1, 1, 64, 64, 32])

Examples using generate_pet_phantom:#

Positron emission tomography (PET) in 2D

Positron emission tomography (PET) in 2D

Positron emission tomography (PET) in 3D

Positron emission tomography (PET) in 3D