KGR introduces continuous k-space representation to boost MRI reconstruction

A new paper from Yu Guan, Qiegen Liu, and colleagues introduces K-space Gaussian Representation (KGR), the first explicit continuous representation formulated directly in the native k-space domain for parallel MRI. Traditional accelerated MRI reconstruction estimates missing k-space samples on discrete grids using interpolation operators or structure priors. KGR instead parameterizes the continuous signal using Gabor-Gaussian primitives with shared spatial geometry, producing a compact representation that naturally preserves inter-coil correlations. Because unconstrained continuous fitting may violate multi-coil signal structure, the authors project the representation onto a low-rank manifold, enforcing algebraic constraints from smoothly varying phase and coil redundancy.

A frequency-adaptive fitting strategy accounts for the heterogeneous characteristics of different k-space regions, further improving reconstruction fidelity. Comprehensive validation across multiple datasets and sampling schemes shows that KGR consistently outperforms representative reconstruction baselines in quantitative metrics and visual quality. The work suggests that explicit continuous parameterization of k-space offers a principled framework for integrating continuous signal modeling with structured low-rank reconstruction, potentially enabling faster MRI scans without sacrificing image quality.