Overlap‐Kernel <scp>EPI</scp> : Estimating <scp>MRI</scp> Shot‐to‐Shot Phase Variations by Shifted‐Kernel Extraction From Overlap Regions at Arbitrary k‐Space Locations

Rui Tian, Martin Uecker, Maxim Zaitsev, Klaus SchefflerMagnetic 2025 · DOI

TL;DR

This paper introduces "Overlap-Kernel EPI," a novel self-navigation technique for MRI that accurately estimates and corrects shot-to-shot phase variations in multi-shot EPI sequences. By leveraging small, overlapping regions in k-space, this method significantly improves image quality, especially in diffusion-weighted scans, without requiring dedicated navigator scans.

What Problem Does This Paper Address?

Multi-shot MRI techniques, like Echo Planar Imaging (EPI), are crucial for achieving higher spatial resolution and reducing geometric distortions. However, these benefits are often undermined by "ghosting artifacts" caused by subtle, shot-to-shot variations in the MRI signal's phase. These phase fluctuations can arise from patient motion, system imperfections, or eddy currents. Traditional methods to correct these artifacts, such as using dedicated navigator scans or relying on oversampled k-space centers, have limitations. Navigator scans increase scan time, and existing self-navigation methods are often restricted to specific k-space regions or trajectories, limiting their applicability, particularly in challenging sequences like diffusion-weighted EPI. This paper aims to overcome these limitations by developing a more flexible and robust self-navigation strategy.

Methods

The core innovation is the "Overlap-Kernel EPI" method, which extends the concept of self-navigation beyond the k-space center. Instead of relying on dedicated navigator data or heavily oversampled regions, this technique utilizes small, intentionally designed overlaps between k-space segments acquired in consecutive shots. These overlapping regions act as "auto-calibrating signal" (ACS) regions.

The authors adapt principles from parallel imaging techniques like GRAPPA and ESPIRIT. They propose an "ESPIRIT-type operation" that extracts "phase fluctuation kernels" or maps from these overlap regions. This process involves constructing a calibration matrix from the overlapping k-space data, followed by singular-value decomposition (SVD) to extract the relevant phase information. The key insight is that these phase variations can be modeled as k-space kernels that are shift-invariant across k-space locations and RF receiver channels, but shift-variant across shots. This allows for the estimation of shot-to-shot phase differences, which are then used to correct the acquired data during image reconstruction. This approach offers greater flexibility in designing multi-shot sequences, as the overlap regions can be placed at arbitrary k-space locations.

Datasets

The study evaluated their method using in vivo EPI scans acquired on standard 3T MRI scanners (Siemens Prismafit and Cima.X). They tested their approach on four different multi-shot EPI sequences: two-shot mosaic, four-shot mosaic, five-shot readout-segmented mosaic, and five-shot phase-interleaved EPI. They also compared their method against reconstructions using separate 2D navigator acquisitions.

Key Insights and Ideas

  • Flexible Self-Navigation: The ability to place k-space overlaps at arbitrary locations, not just the center, is a significant departure from previous self-navigation methods, greatly increasing sequence design flexibility.
  • Leveraging Parallel Imaging Principles: Adapting GRAPPA/ESPIRIT's kernel-based calibration to estimate phase fluctuations, rather than just coil sensitivities, provides a robust framework.
  • Unified Framework: The paper bridges the gap between RF sensitivity calibration in parallel imaging and signal fluctuation calibration in multi-shot EPI, suggesting a more unified understanding of MRI signal modeling.
  • Robustness to Eddy Currents: The method shows promise in mitigating shot-dependent eddy currents, which are a common challenge in EPI.
  • Improved SNR in Certain Sequences: For specific sequences like two-shot and four-shot mosaic EPI, the self-navigation approach achieved higher SNR compared to navigator-corrected images, likely due to shorter echo times.

Key Results

  • The Overlap-Kernel EPI method successfully corrected shot-to-shot phase variations across various multi-shot EPI trajectories.
  • Reconstructed images using self-navigation showed comparable or even superior quality to those corrected with traditional 2D navigator scans, particularly in terms of ghosting artifacts.
  • Quantitative measurements (coefficient-of-variation maps) indicated that the self-navigation approach often resulted in lower errors and higher SNR, especially in two-shot and four-shot mosaic sequences.
  • The method demonstrated its utility in diffusion-weighted imaging (DWI) and diffusion tensor imaging (DTI), effectively removing artifacts and improving image clarity.
  • The authors showed that their method can be applied to both diffusion-weighted and gradient-echo EPI sequences.

Limitations and Future Work

  • SNR Dependency: The accuracy of kernel extraction can degrade in very low-SNR calibration regions, particularly in peripheral k-space areas of readout-segmented sequences with strong diffusion-weighting. This is an inherent limitation of using low-SNR data for calibration.
  • Phase Wrapping: Phase wrapping can occur, making direct comparison of phase maps from different algorithms challenging.
  • Within-Shot Trajectory Errors: Strong and continuous eddy currents within a single shot can still disrupt the kernel interpolation model, potentially leading to estimation errors. The authors suggest that incorporating more realistic within-shot trajectory corrections could improve results.
  • Future Work: The authors suggest exploring higher b-values, utilizing high-performance gradient scanners, and investigating neural network-based kernel extraction for improved accuracy. They also propose extending the method to 3D volumetric scans and simultaneous multi-slice (SMS) imaging. Further research into optimizing overlap regions and kernel sizes for specific applications is also recommended.

Summary updated

Abstract

This paper rigorously analyzes subspace algorithms for phase-stabilized multi-shot MRI, significantly broadens the self-navigation concept, and provides a more complete picture of the connections between signal fluctuations calibration in EPI and RF sensitivity calibration in parallel imaging.
BibTeX
@inproceedings{tian2025overlapkernel,
  author = {Rui Tian and Martin Uecker and Maxim Zaitsev and Klaus Scheffler},
  title = {{Overlap‐Kernel <scp>EPI</scp> : Estimating <scp>MRI</scp> Shot‐to‐Shot Phase Variations by Shifted‐Kernel Extraction From Overlap Regions at Arbitrary k‐Space Locations}},
  booktitle = {Magnetic},
  year = {2025},
  doi = {10.1002/mrm.70196},
}
Figure