Weakly Convex Ridge Regularization for 3D Non-Cartesian MRI Reconstruction.
Authors
Affiliations (4)
Affiliations (4)
- Faculty of Mathematics, Chemnitz University of Technology, Chemnitz, Germany.
- NeuroSpin, Frédéric Joliot Institute for Life Sciences, CEA Paris-Saclay, Gif-sur-Yvette, France.
- MIND, Inria Saclay Centre, Palaiseau, France.
- Siemens Healthineers, Courbevoie, France.
Abstract
While highly accelerated non-Cartesian acquisition protocols significantly reduce scan time, they often entail long reconstruction delays for 3D data. Deep learning-based reconstruction methods can alleviate this, but typically rely on complex, highly parametrized architectures. As an alternative, we introduce a rotation-invariant weakly convex ridge regularizer (WCRR). The resulting variational reconstruction approach is benchmarked against state-of-the-art methods on retrospectively simulated data and out-of-distribution prospective GoLF-SPARKLING and CAIPIRINHA acquisitions. Our approach achieves competitive reconstruction quality while using orders of magnitude fewer parameters and retaining the interpretability of a sparsity-promoting variational prior. In summary, WCRR provides a compact and interpretable learned variational prior for 3D non-Cartesian MRI reconstruction.