Gabor Primitives for Accelerated Cardiac Cine MRI Reconstruction

Wenqi Huang; Veronika Spieker; Nil Stolt-Ansó; Natascha Niessen; Maik Dannecker; Sevgi Gokce Kafali; Sila Kurugol; Julia A. Schnabel

eess.IVMar 5, 2026

Gabor Primitives for Accelerated Cardiac Cine MRI Reconstruction

Authors: Wenqi Huang, Veronika Spieker, Nil Stolt-Ansó, Natascha Niessen, Maik Dannecker, Sevgi Gokce Kafali, Sila Kurugol, Julia A. Schnabel, +1 more

Organizations: Chair for AI in Healthcare and Medicine, Technical University of Munich (TUM) and TUM University Hospital, Munich, Germany · School of Computation and Information Technology, Technical University of Munich, Munich, Germany · Institute of Machine Learning in Biomedical Imaging, Helmholtz Munich, Neuherberg, Germany · Department of Radiology, Boston Children’s Hospital and Harvard Medical School, Boston, USA · Munich Center for Machine Learning, Technical University of Munich, Munich, Germany · GE HealthCare, Munich, Germany · School of Biomedical Engineering and Imaging Sciences, King’s College London, London, United Kingdom · Department of Computing, Imperial College London, London, United Kingdom

Abstract

Accelerated cardiac cine MRI requires reconstructing spatiotemporal images from highly undersampled k-space data. Implicit neural representations (INRs) enable scan-specific reconstruction without large training datasets, but encode content implicitly in network weights without physically interpretable parameters. Gaussian primitives provide an explicit and geometrically interpretable alternative, but their spectra are confined near the k-space origin, limiting high-frequency representation. We propose Gabor primitives for MRI reconstruction, modulating each Gaussian envelope with a complex exponential to place its spectral support at an arbitrary k-space location, enabling efficient representation of both smooth structures and sharp boundaries. To exploit spatiotemporal redundancy in cardiac cine, we decompose per-primitive temporal variation into a low-rank geometry basis capturing cardiac motion and a signal-intensity basis modeling contrast changes. Experiments on cardiac cine data with Cartesian and radial trajectories show that Gabor primitives consistently outperform compressed sensing, Gaussian primitives, and hash-grid INR baselines, while providing a compact, continuous-resolution representation with physically meaningful parameters.

Figures & tables

Figure 1: Left: Gaussian vs. Gabor primitives in image space and k -space. A Gaussian’s spectral support is fixed at the k -space origin; a Gabor primitive shifts it to $\vec{\xi}_{i}$ via complex-exponential modulation, enabling more efficient frequency coverage. Right: Cardiac cine image is modeled as a mixture of time-varying Gabor primitives. Geometry parameters ( $\vec{\mu}$ , $\vec{s}$ , $\theta$ , $\vec{\xi}$ ) and complex weights $w$ are generated from low-rank geometry (blue) and intensity (orange) bases. Primitives are rasterized, passed through a multi-coil forward model, and fitted to acquired k -space data end-to-end.

		Cartesian $R{=}12$				Cartesian $R{=}16$				Radial $R{\approx}23$
Method	$\rho\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	FSIM $\uparrow$	Time $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	FSIM $\uparrow$	Time $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	FSIM $\uparrow$	Time $\downarrow$
L+S [ 16 ]	—	37.15	0.940	0.850	0.3	32.19	0.824	0.777	0.4	35.82	0.922	0.810	1.5
PICS [ 14 ]	—	42.21	0.972	0.886	0.1	41.05	0.966	0.875	0.1	35.19	0.905	0.793	1.1
Hash-INR [ 15 ]	2.60	39.63	0.921	0.841	4.6	37.77	0.893	0.815	4.6	34.00	0.849	0.767	8.0
Gaussian	0.42	41.50	0.972	0.874	1.2	40.67	0.967	0.864	1.2	36.67	0.942	0.810	3.4
Gabor (ours)	0.41	42.61	0.974	0.883	2.5	41.39	0.968	0.871	2.5	37.53	0.947	0.830	4.2

Table 1: Quantitative comparison on cardiac cine MRI (

N{=}99

Cartesian,

N{=}102

radial). PSNR in dB, Time in minutes.

\rho=N_{\text{params}}/(2HWT)

: ratio of learnable parameters to real-valued image degrees of freedom. Best in bold , second best underlined .

Figure 2: Reconstruction comparison on Cartesian (top) and radial (bottom) cardiac cine. Each block shows a representative frame with metrics, zoomed ROI with error map, and

y

–

t

profile with temporal error.

Figure 3: Unique capabilities of Gabor primitives. (a) k -space primitive distribution: each dot is a primitive’s center frequency $\vec{\xi}_{n}$ with $3\sigma$ support ellipses. (b) Frequency-band PSNR (low, mid, high) on the radial dataset, averaged over all subjects. (c) Frequency decomposition: primitives partitioned by $|\vec{\xi}_{n}|$ into low- ( $<\!\tfrac{1}{4}$ ) and high-frequency ( $\geq\!\tfrac{1}{4}$ ) groups and their sum. (d) $4\times$ super-resolution on the marked ROI; bicubic upsampling of the reference shown for comparison.

Explore similar work

Primitive Representation Learning for Unsupervised Dynamic Contrast Enhanced MRI Reconstruction
Aug 18, 2026Veronika Spieker, Wenqi Huang, Cemre Ariyurek +5Magnetic Resonance Imaging Accelerated MRI Reconstruction
Institute of Machine Learning in Biomedical Imaging, Helmholtz Munich, Germany · School of Computation and Information Technology, Technical University of Munich, Germany · Department of Radiology, Boston Children’s Hospital and Harvard Medical School, USA +3
Enhancing Implicit Neural Representations with Image Feature Embedding for Unsupervised Cardiac Cine MRI Reconstruction
Jul 5, 2026Donghang Lyu, Marius Staring, Yiming Dong +3Implicit Neural Representations Accelerated MRI Reconstruction
Division of Image Processing, Department of Radiology, Leiden University Medical Center, 2333ZA Leiden, Netherlands · C.J. Gorter Center, Department of Radiology, Leiden University Medical Center, 2333ZA Leiden, Netherlands · Philips Innovative Technologies, 22335 Hamburg, Germany +1
K-space Gaussian Representation for Parallel MRI
Jul 29, 2026Yu Guan, Mingyu Hu, Jiale Hu +3Accelerated MRI Reconstruction Magnetic Resonance Imaging
School of Advanced Manufacturing and the School of Information Engineering, Nanchang University, Nanchang 330031, China · Key Laboratory of Biomedical Imaging Science and System, Chinese Academy of Sciences, Shenzhen 518055, China · School of Information Engineering, Nanchang University, Nanchang 330031, China +1

Aug 18, 2026J/K move · Enter open · S save

Primitive Representation Learning for Unsupervised Dynamic Contrast Enhanced MRI Reconstruction

Veronika Spieker, Wenqi Huang, Cemre Ariyurek, Liam Timms+4

Institute of Machine Learning in Biomedical Imaging, Helmholtz Munich, Germany · School of Computation and Information Technology, Technical University of Munich, Germany · Department of Radiology, Boston Children’s Hospital and Harvard Medical School, USA+3

Magnetic Resonance Imaging Accelerated MRI Reconstruction Representation Learning

Open paper AskPDF arXiv

Reliable quantitative analysis of dynamic contrast-enhanced MRI requires high-quality spatiotemporal reconstructions at high undersampling rates. Scan-specific reconstructions using Gaussian and Gabor primitives have shown promising results without the need for large training datasets, but have not addressed the additional dimension of dynamic contrast. We propose a multi-dimensional, primitive based framework for dynamic contrast-enhanced MRI reconstruction that disentangles the underlying anatomy, the dynamic contrast enhancement, and residual motion into separate temporal basis functions, thereby enabling a geometrical interpretation of the representation. We show that this architecture achieves performance competitive with conventional reconstruction methods, both in reconstruction quality and in the accuracy of extracted aorta and kidney enhancement curves. The modular tier design extends naturally to additional dynamic factors and higher acceleration rates. Code available at https://github.com/compai-lab/2026-GaborDCE-spieker.

Abstract

Explore similar work

Primitive Representation Learning for Unsupervised Dynamic Contrast Enhanced MRI Reconstruction

Enhancing Implicit Neural Representations with Image Feature Embedding for Unsupervised Cardiac Cine MRI Reconstruction

K-space Gaussian Representation for Parallel MRI