Gabor Primitives for Accelerated Cardiac Cine MRI Reconstruction

A collaborative research team from institutions including Imperial College London and the Technical University of Munich has introduced a novel AI-driven method for reconstructing cardiac cine MRI scans. Published on arXiv, the paper "Gabor Primitives for Accelerated Cardiac Cine MRI Reconstruction" addresses a critical bottleneck in medical imaging: the need to create clear, spatiotemporal images from highly undersampled k-space data to speed up scan times. The team's innovation replaces traditional implicit neural representations (INRs)—which encode image content opaquely within network weights—with explicit, geometrically interpretable "Gabor primitives."

These primitives work by modulating a standard Gaussian envelope with a complex exponential. This key modification allows the spectral support of each primitive to be positioned at an arbitrary location in k-space, enabling the efficient representation of both the smooth anatomical structures and the sharp boundaries essential for accurate cardiac diagnosis. To handle the dynamic nature of a beating heart, the model cleverly decomposes the temporal variation of each primitive into two low-rank bases: one capturing the underlying cardiac motion and another modeling changes in signal intensity or contrast.

In practical testing on cardiac cine data using both Cartesian and radial sampling trajectories, the Gabor primitives method demonstrated superior performance. It consistently outperformed established benchmarks, including conventional compressed sensing techniques, simpler Gaussian primitives, and modern hash-grid INR baselines. Beyond just higher fidelity reconstructions, the approach provides a continuous-resolution representation of the scan, meaning it's not locked to a fixed pixel grid, and yields a set of physically meaningful parameters that radiologists and researchers can potentially analyze directly, moving towards more interpretable AI in medicine.