Magnetic Resonance Imaging (MRI) stands as a cornerstone of modern diagnostic medicine, offering unparalleled soft-tissue contrast without the use of ionizing radiation. For decades, the standard approach to data acquisition in MRI has relied on Cartesian trajectories: a point-by-point fill of k-space along a rectilinear grid. This method, while robust and conceptually simple, imposes fundamental trade-offs between scan time, resolution, and motion sensitivity. In recent years, non-Cartesian trajectories have emerged as a powerful alternative, enabling faster acquisitions, improved image quality, and novel imaging strategies that push the boundaries of clinical MRI. Understanding the physics behind these non-rectilinear pathways is essential for radiologists, physicists, and engineers seeking to leverage their full potential.

What Are Non-Cartesian MRI Trajectories?

In MRI, the raw data that ultimately forms an image is collected in the spatial frequency domain, known as k-space. A Cartesian trajectory samples k-space row by row, with each line acquired after a separate excitation. Non-Cartesian trajectories, by contrast, do not follow a regular grid. Instead, they traverse k-space along paths such as radial lines, spirals, rosettes, or even pseudo-random zigzags. Each trajectory type offers distinct sampling properties that can be tailored to specific imaging goals.

  • Radial trajectories sample k-space along lines that emanate from the center outward, akin to spokes on a wheel. Over-sampling of the center of k-space provides inherent motion robustness and allows for retrospective correction of subject movement.
  • Spiral trajectories begin at the center of k-space and unwind outward in a continuous curve. Their high sampling efficiency, especially for dynamic or volumetric imaging, reduces scan time dramatically while maintaining a high signal-to-noise ratio (SNR).
  • Rosette and Lissajous trajectories produce self-intersecting patterns that can encode very high spatial frequencies in a single shot, useful for ultra-fast imaging.
  • Zigzag or “boustrophedon” paths are sometimes used in combination with parallel imaging to accelerate acquisitions further.

Each of these patterns imposes different requirements on gradient hardware, reconstruction algorithms, and sequence design. The choice of trajectory depends on the clinical application, the region of interest, and the acceptable trade-off between speed and image quality.

The Physics Behind Non-Cartesian Trajectories

To understand how non-Cartesian trajectories work, one must revisit the fundamental relationship between magnetic field gradients and the location of data in k-space. The MR signal equation in the presence of time-varying gradients can be written as:

S(t) = ∫ ρ(𝐫) exp(−i2π 𝐤(t) · 𝐫) d𝐫

where 𝐤(t) is the time-integral of the gradient waveform 𝐆(t), scaled by the gyromagnetic ratio γ:

𝐤(t) = γ ∫₀ᵗ 𝐆(τ) dτ

In Cartesian imaging, gradients are constant during the readout, creating a linear relationship between time and k-space position. Non-Cartesian trajectories require time-varying gradient waveforms that trace a path through k-space according to the desired geometry. For example, a spiral trajectory demands sinusoidal gradients that simultaneously excite the x- and y-axes, with the amplitude and frequency carefully shaped to produce a smooth spiral without violating hardware slew-rate constraints.

Gradient Design and Waveform Shaping

The design of gradient waveforms for non-Cartesian trajectories is a non-trivial optimization problem. Gradient systems have maximum amplitude (typically 30–80 mT/m on clinical scanners) and maximum slew rate (usually 100–200 mT/m/ms). For a spiral trajectory, the gradient waveform must accelerate quickly as the spiral moves outward, yet remain within allowable limits to avoid peripheral nerve stimulation or scanner overheating. Advanced trajectory design uses time-optimal solutions, often computed via numerical methods, to minimize the total readout duration while respecting hardware constraints. The resulting gradient waveforms are then pre-emphasized to compensate for eddy-current effects, which can distort the intended path. Small errors in gradient fidelity can lead to image artifacts such as blurring or signal dropout, making precise calibration essential.

Signal Acquisition and Sampling Density

Non-Cartesian trajectories sample k-space non-uniformly. Radial trajectories, for instance, collect more samples near the center (low spatial frequencies) and fewer at the periphery (high frequencies). This variable density influences the SNR and the artifact behavior. The Nyquist criterion must be satisfied for the highest spatial frequency of interest; otherwise, aliasing (streak artifacts in radial imaging or spiral blurring) can occur. Sampling density also affects reconstruction: because data points are not on a Cartesian grid, they cannot be directly transformed via the fast Fourier transform (FFT). Instead, reconstruction must account for the irregular positions, typically using gridding algorithms (e.g., Kaiser-Bessel-based convolution) or iterative methods that solve an inverse problem. Advanced reconstruction may also incorporate coil sensitivity maps (for parallel imaging) and regularization terms to suppress noise and artifacts.

Key Benefits Over Cartesian Trajectories

Non-Cartesian trajectories offer several distinct advantages that complement or exceed conventional Cartesian methods. These benefits have driven their adoption in both research and clinical settings.

Faster Imaging and Reduced Scan Times

One of the most compelling benefits is speed. Radial and spiral trajectories can cover k-space in a single shot, acquiring a whole image with just one excitation (or a small number of interleaves). For dynamic studies—such as cardiac cine imaging, contrast-enhanced perfusion, or functional MRI—this rapid acquisition captures physiological changes with higher temporal resolution. In extreme cases, spiral imaging can achieve sub-second frame rates, enabling real-time monitoring of swallowing, joint motion, or fetal movements. Even when multi-shot (multiple excitations) are used to improve image quality, non-Cartesian methods often require fewer shots than Cartesian phase-encoding to achieve the same resolution, directly cutting scan time.

Motion Robustness

Motion artifacts remain a major source of image degradation in clinical MRI, especially for uncooperative patients, pediatric scans, or respiratory-gated abdomen examinations. Non-Cartesian trajectories, particularly radial, are inherently less sensitive to motion. Over-sampling of the k-space center means that even if a few radial spokes are corrupted by motion, the overall image retains most of its low-frequency content, and the artifacts tend to appear as mild blurring rather than ghosting. Retrospective motion correction techniques can exploit the central redundancy to estimate and compensate for rigid-body motion between spokes. Similarly, spiral trajectories can be designed to be “self-navigated”: the central region of k-space is sampled repeatedly, allowing direct extraction of motion parameters during reconstruction.

Higher Resolution and Flexibility

Non-Cartesian trajectories can be engineered to achieve high isotropic resolution in three dimensions without prohibitively long scan times. For instance, a stack-of-stars radial acquisition (radial in-plane, Cartesian through-plane) is used for high-resolution whole-brain imaging and dynamic contrast-enhanced MRI of the breast. Spiral fan-beam sampling can provide sub-millimeter resolution for neuroimaging. Moreover, these trajectories permit partial Fourier or sparse sampling strategies—such as compressed sensing—that Cartesian trajectories do not accommodate as naturally. By combining non-Cartesian sampling with iterative reconstruction, clinicians can trade speed for SNR or vice versa, adapting the protocol to each patient’s needs.

Clinical Applications

Non-Cartesian MRI trajectories have moved from research prototypes to routine clinical use in several domains. Their ability to deliver high-quality images quickly makes them indispensable in time-sensitive scenarios.

  • Cardiac MRI: Spiral and radial sequences are used for cine imaging (heart wall motion), myocardial perfusion (first-pass contrast dynamics), and T1/T2 mapping. The speed of spiral readouts enables breath-hold acquisitions of less than 10 seconds. Radial trajectories reduce artifacts from breathing and cardiac motion without the need for electrocardiogram (ECG) gating in some cases.
  • Neuroimaging: Radial imaging is employed in arterial spin labeling (ASL) brain perfusion, where the inherent motion robustness improves the quantification of cerebral blood flow. Spiral trajectories are the backbone of many functional MRI (fMRI) protocols because they achieve whole-brain coverage with high temporal resolution for detecting blood-oxygen-level-dependent (BOLD) signal changes.
  • Abdominal and Pelvic Imaging: Free-breathing radial sequences (e.g., RARE or GRE variants) allow for robust liver and kidney imaging without breath-holding—beneficial for patients unable to follow commands. The reduced sensitivity to respiratory motion also improves image quality in pediatric and obese patients.
  • Musculoskeletal Imaging: Ultrashort echo time (UTE) sequences using radial sampling enable the visualization of tissues with very short T2 relaxation times, such as cortical bone, ligaments, and tendons. These structures appear dark in conventional Cartesian sequences but become visible with UTE-radial methods.

Beyond these applications, non-Cartesian trajectories are being explored for hyperpolarized 13C imaging, MR spectroscopy, and simultaneous multislice (SMS) acquisitions where rapid excitation and readout are critical.

Challenges and Considerations

While the benefits are substantial, non-Cartesian MRI is not a panacea. Several technical hurdles must be addressed for widespread deployment.

Reconstruction complexity: Non-uniform sampling demands specialized reconstruction algorithms that are computationally intensive. Gridding-based methods are fast but can introduce artifacts if density compensation functions are inaccurately computed. Iterative reconstructions (e.g., using total variation or wavelet regularization) yield higher image quality but require significant processing time, often offloaded to GPU accelerators in clinical workflows. The reliance on non-standard reconstruction also means that compatibility with existing PACS and image processing pipelines may require additional steps.

Hardware limitations: The rapid gradient switching required for spiral or rosette trajectories can exceed the slew rate of many clinical systems, limiting the achievable resolution or field of view. Eddy currents and gradient delays cause trajectory deviations that must be corrected through calibration. With the advent of higher-performance gradients (such as those in the new generation of whole-body 5T and 7T systems), these constraints are gradually being relaxed, but careful sequence design remains necessary.

Susceptibility and off-resonance effects: Non-Cartesian trajectories, particularly spirals, are sensitive to magnetic field inhomogeneities and chemical shift. Off-resonance frequences cause blurring or distortions that are not easily corrected. Advanced techniques such as frequency-segmented reconstruction or field-map-based correction can mitigate these artifacts, but they add to the computational burden. Radial trajectories are less affected by off-resonance but still require correction for eddy currents and gradient delays.

Standardization and regulatory pathways: Most non-Cartesian sequences are developed as vendor-specific “work-in-progress” packages. Achieving uniform quality across different scanner manufacturers and software versions remains a challenge. The FDA and other regulatory agencies must approve each new sequence variant, which can slow clinical adoption. Nevertheless, the growing body of evidence for improved patient outcomes—especially in pediatric and emergency imaging—is driving manufacturers to include more non-Cartesian options in their commercial packages.

Future Directions

The evolution of non-Cartesian MRI trajectories is far from complete. Emerging technologies and methodologies promise to extend their capabilities even further.

  • Artificial intelligence (AI) in reconstruction: Deep learning models, such as convolutional neural networks and generative adversarial networks, can perform rapid trajectory estimation and artifact removal. AI-based “autoreconstruction” from non-Cartesian data may eliminate the need for explicit gridding, enabling real-time imaging. Early results show that neural nets can learn to correct motion and off-resonance effects on the fly.
  • Adaptive and compressed sensing: Non-Cartesian trajectories are a natural fit for compressed sensing, which exploits sparsity in the image domain to allow sub-Nyquist sampling. Adaptive trajectories that concentrate sampling in regions of high image gradient (as determined by a real-time reconstruction) can further accelerate acquisitions without sacrificing quality.
  • Combined with magnetic resonance fingerprinting (MRF): MRF uses rapidly varying acquisition parameters to simultaneously quantify multiple tissue properties (T1, T2, proton density). Non-Cartesian sampling is critical for achieving the necessary acceleration in MRF, and radial or spiral variants are now standard in research implementations. This approach is moving toward clinical translation for whole-brain quantitative mapping.
  • Low-field and portable MRI: The interest in compact, low-field (e.g., 0.05–0.1 T) MRI scanners for point-of-care imaging has revived interest in non-Cartesian trajectories. At low field, SNR is inherently lower, and scan time is even more precious. Radial and spiral sequences can maximize data efficiency while reducing the impact of motion in uncooperative patients.

Leading research institutions now routinely publish results from non-Cartesian sequences that approach four-dimensional (4D) imaging—capturing three spatial dimensions plus time in pediatric respiration, freely moving volunteers, or dynamic contrast studies of the brain. As reconstruction hardware accelerates and algorithms mature, these methods are expected to become the default for many clinical protocols within the next decade.

Conclusion

Non-Cartesian MRI trajectories represent a paradigm shift from the rigid grid of conventional Cartesian acquisition. By embracing the physics of gradient design and leveraging advanced reconstruction, these trajectories offer faster imaging, greater robustness to motion, and flexibility to adapt to a wide range of clinical needs. While challenges such as reconstruction complexity and hardware constraints remain, ongoing advances in computational power, AI, and gradient technology are steadily lowering the barriers to adoption. For radiologists and researchers, understanding the principles behind radial, spiral, and other non-Cartesian paths is essential for selecting the right sequence for each patient and for interpreting the resultant images accurately. As the field progresses, non-Cartesian methods are poised to become not merely an alternative, but a cornerstone of modern MRI practice.