Phase2PhaseSplittingMaskGenerator

class deepinv.physics.generator.Phase2PhaseSplittingMaskGenerator(img_size, device='cpu', rng=None)

Bases: BernoulliSplittingMaskGenerator

Phase2Phase splitting mask generator for dynamic data.

To be exclusively used with deepinv.loss.mri.Phase2PhaseLoss. Splits dynamic data (i.e. data of shape (B, C, T, H, W)) into even and odd phases in the T dimension.

Used in Eldeniz et al.[1].

If input_mask not passed, a blank input mask is used instead.

Parameters:

• img_size (tuple[int]) – size of the tensor to be masked without batch dimension of shape (C, T, H, W). Note this can be overriden on-the-fly by passing in img_size or input_mask arguments to step.

• device (str, torch.device) – device where the tensor is stored (default: ‘cpu’).

• rng (torch.Generator) – unused.

---

References:

[1]

Cihat Eldeniz, Weijie Gan, Sihao Chen, Tyler J Fraum, Daniel R Ludwig, Yan Yan, Jiaming Liu, Thomas Vahle, Uday Krishnamurthy, Ulugbek S Kamilov, and others. Phase2phase: respiratory motion-resolved reconstruction of free-breathing magnetic resonance imaging using deep learning without a ground truth for improved liver imaging. Investigative Radiology, 56(12):809–819, 2021.

batch_step(input_mask=None, img_size=None)

Create one batch of splitting mask.

Parameters:

• input_mask (torch.Tensor, None) – optional mask to be split. If None, all pixels are considered. If not None, only pixels where mask==1 are considered. Batch dimension should not be included in shape.

• img_size (tuple) – if not None, generate masks of this 2D image shape and override img_size attribute, must be of form (H, W).

Returns:

mask without batch dimension of shape specified either by img_size, input_mask, or class attribute img_size.

Return type:

dict