mr2.algorithms.rovir

mr2.algorithms.rovir(img: Tensor, roi_mask: Tensor, *, n_compressed_coils: int) → EinsumOp[source]

Orthonormal ROVir coil compression operator.

ROVir emphasizes signal variation inside an ROI while suppressing a background region. This implementation estimates ROI and background covariances, solves the generalized eigenproblem via background whitening, and keeps the dominant eigenvectors.

An additional QR step produces an orthonormal compression matrix. This differs from the original formulation [KIM2021], which uses the generalized eigenvectors directly (they are background-orthogonal, not Euclidean- orthonormal). The QR step preserves the compressed subspace but loses the individual signal-to-interference interpretation and the strict eigenvalue ordering. The orthonormal basis is convenient for prewhitened input because it leaves noise statistics unchanged.

Parameters:

  • img (Tensor) – Prewhitened coil images, shape (coil, z, y, x).

  • roi_mask (Tensor) – Boolean mask of shape (z, y, x) selecting the ROI. The background is defined as the complement of roi_mask.

  • n_compressed_coils (int) – Number of virtual coils to retain.

Returns:

Operator (..., coil, z, y, x) -> (..., compressed_coil, z, y, x).

References

[KIM2021]

Kim D, Cauley SF, Nayak KS, Leahy RM, Haldar JP.

Region-optimized virtual (ROVir) coils: Localization and/or suppression of spatial regions using sensor-domain beamforming. Magn Reson Med. 2021;86(1):197–212. https://doi.org/10.1002/mrm.28706