Image Reconstruction with Low-rankness and Self-consistency of k-space Data in Parallel MRI

Xinlin Zhang, Di Guo, Yiman Huang, Ying Chen, Liansheng Wang, Feng Huang, Xiaobo Qu

Abstract

Parallel magnetic resonance imaging has served as an effective and widely adopted technique for accelerating scans. The advent of sparse sampling offers aggressive acceleration, allowing flexible sampling and better reconstruction. Nevertheless, faithfully reconstructing the image from limited data still poses a challenging task. Recent low-rank reconstruction methods exhibit superiority in providing a high-quality image. However, none of them employ the routinely acquired calibration data for improving image quality in parallel magnetic resonance imaging. In this work, an image reconstruction approach named STDLR-SPIRiT was proposed to explore the simultaneous two-directional low-rankness (STDLR) in the k-space data and to mine the data correlation from multiple receiver coils with the iterative self-consistent parallel imaging reconstruction (SPIRiT). The reconstruction problem was then solved with a singular value decomposition-free numerical algorithm. Experimental results of phantom and brain imaging data show that the proposed method outperforms the state-of-the-art methods in terms of suppressing artifacts and achieving the lowest error. Moreover, the proposed method exhibits robust reconstruction even when the auto-calibration signals are limited in parallel imaging. Overall the proposed method can be exploited to achieve better image quality for accelerated parallel magnetic resonance imaging.

Introduction

Magnetic Resonance Imaging (MRI) is a critical diagnostic tool, yet its clinical utility is often hampered by slow acquisition speeds. Parallel Imaging (pMRI) and sparse sampling are standard techniques used to accelerate scans by collecting only a fraction of the required k-space data. The primary challenge in pMRI is the accurate reconstruction of missing data from these undersampled acquisitions.

Current reconstruction strategies generally fall into two categories: calibration-based methods like SPIRiT, which utilize Auto-Calibration Signals (ACS) to learn coil correlations, and low-rank structured matrix methods like ALOHA, which exploit the inherent low-rank properties of k-space data. However, calibration-based methods often fail when ACS data is limited, and existing low-rank methods typically process data directions sequentially, leading to suboptimal artifact suppression. This paper introduces STDLR-SPIRiT, a framework that bridges these two approaches by enforcing simultaneous two-directional low-rankness alongside coil self-consistency.

Methodology: The STDLR-SPIRiT Framework

The proposed STDLR-SPIRiT model integrates three core innovations to improve reconstruction quality:

  1. Simultaneous Two-Directional Low-Rankness (STDLR): Unlike sequential approaches that can propagate errors between horizontal and vertical processing passes, STDLR enforces low-rank constraints in both directions simultaneously. This joint optimization effectively suppresses ringing artifacts.
  2. SPIRiT Integration: By incorporating a self-consistency term from the SPIRiT framework, the model ensures the reconstructed data adheres to the local linear relationships between multiple receiver coils, leveraging the physical redundancy of the coil array.
  3. SVD-free Optimization: To address the high computational cost of Singular Value Decomposition (SVD) in standard low-rank optimization, the authors employ a matrix factorization technique within an Alternating Direction Method of Multipliers (ADMM) framework. This allows for efficient reconstruction of large datasets.
Overview of the STDLR-SPIRiT reconstruction framework.

Experimental Evaluation

The performance of STDLR-SPIRiT was evaluated using phantom data (3T Siemens Trio) and human brain data (T2-weighted and T1-weighted) across various scanner types (Siemens, Philips, GE). The researchers tested the method against several sampling patterns, including Cartesian, pseudo-radial, and 2D random undersampling.

Results and Discussion

Experimental results demonstrate that STDLR-SPIRiT consistently outperforms state-of-the-art methods such as GRAPPA, $\ell_1$-SPIRiT, and ALOHA. Quantitative metrics, including Relative $\ell_2$ Norm Error (RLNE) and Mean Structural Similarity (MSSIM), were significantly improved.

A key finding is the method's robustness to limited calibration data. While standard SPIRiT performance degrades sharply as the number of ACS lines decreases, STDLR-SPIRiT maintains high reconstruction fidelity even with minimal calibration signals.

Error maps and performance comparison against varying ACS lines.

Qualitatively, the proposed method preserves sharp edges and fine structural details that are often blurred or obscured by noise in competing methods. The synergy between the low-rank mathematical prior and the data-driven coil consistency provides a powerful mechanism for high-quality pMRI reconstruction.

Conclusion

STDLR-SPIRiT offers a robust and effective solution for accelerated MRI reconstruction. By combining simultaneous directional low-rankness with multi-coil self-consistency and utilizing an efficient SVD-free algorithm, it achieves superior image quality and artifact suppression. Future work will focus on further reducing computation time through GPU acceleration and extending the framework to 3D and dynamic MRI applications.

Summary updated

Abstract

Parallel magnetic resonance imaging has served as an effective and widely adopted technique for accelerating scans. The advent of sparse sampling offers aggressive acceleration, allowing flexible sampling and better reconstruction. Nevertheless, faithfully reconstructing the image from limited data still poses a challenging task. Recent low-rank reconstruction methods exhibit superiority in providing a high-quality image. However, none of them employ the routinely acquired calibration data for improving image quality in parallel magnetic resonance imaging. In this work, an image reconstruction approach named STDLR-SPIRiT was proposed to explore the simultaneous two-directional low-rankness (STDLR) in the k-space data and to mine the data correlation from multiple receiver coils with the iterative self-consistent parallel imaging reconstruction (SPIRiT). The reconstruction problem was then solved with a singular value decomposition-free numerical algorithm. Experimental results of phantom and brain imaging data show that the proposed method outperforms the state-of-the-art methods in terms of suppressing artifacts and achieving the lowest error. Moreover, the proposed method exhibits robust reconstruction even when the auto-calibration signals are limited in parallel imaging. Overall the proposed method can be exploited to achieve better image quality for accelerated parallel magnetic resonance imaging.
BibTeX
@article{zhang2020image,
  author = {Xinlin Zhang and Di Guo and Yiman Huang and Ying Chen and Liansheng Wang and Feng Huang and Xiaobo Qu},
  title = {{Image Reconstruction with Low-rankness and Self-consistency of k-space Data in Parallel MRI}},
  journal = {Medical Image Analysis},
  year = {2020},
  url = {https://openreview.net/forum?id=TXlN7f6_XzXP},
}
Figure