Shearlet-based compressed sensing for fast 3D cardiac MR imaging using iterative reweighting

Jackie Ma, Maximilian März, Stephanie Funk, Jeanette Schulz-Menger, Gitta Kutyniok, Tobias Schaeffter, Christoph KolbitscharXiv 2017 · DOI · OpenReview

Abstract: High-resolution three-dimensional (3D) cardiovascular magnetic resonance (CMR) is a valuable medical imaging technique, but its widespread application in clinical practice is hampered by long acquisition times. Here we present a novel compressed sensing (CS) reconstruction approach using shearlets as a sparsifying transform allowing for fast 3D CMR (3DShearCS). Shearlets are mathematically optimal for a simplified model of natural images and have been proven to be more efficient than classical systems such as wavelets. Data is acquired with a 3D Radial Phase Encoding (RPE) trajectory and an iterative reweighting scheme is used during image reconstruction to ensure fast convergence and high image quality. In our in-vivo cardiac MRI experiments we show that the proposed method 3DShearCS has lower relative errors and higher structural similarity compared to the other reconstruction techniques especially for high undersampling factors, i.e. short scan times. In this paper, we further show that 3DShearCS provides improved depiction of cardiac anatomy (measured by assessing the sharpness of coronary arteries) and two clinical experts qualitatively analyzed the image quality.

Introduction

High-resolution 3D Cardiovascular Magnetic Resonance (CMR) is the gold standard for visualizing complex heart anatomy and coronary arteries. However, it suffers from a major clinical bottleneck: acquisition time. Because the heart is constantly moving and the patient is breathing, data must be captured in specific windows, leading to scan times often exceeding 15 minutes. To accelerate this process, researchers use Compressed Sensing (CS), which reconstructs full images from undersampled data. The success of CS depends on a "sparsifying transform"; while wavelets are traditionally used, they struggle to efficiently represent the curved edges and fine vessels typical of cardiac anatomy.

Methodology: The 3DShearCS Framework

The authors propose 3DShearCS, a framework combining three key components to overcome traditional limitations:

  1. 3D Shearlet Transform: Unlike wavelets, which use isotropic scaling, shearlets use anisotropic scaling and shearing. This allows them to represent smooth areas bounded by curved edges with far fewer coefficients, making them mathematically optimal for medical images.
  2. Iterative Reweighting: To enhance sparsity and speed up convergence, the algorithm adaptively changes the weights of shearlet coefficients during reconstruction, helping the solver distinguish true anatomy from noise.
  3. Non-Cartesian ADMM Solver: Data is acquired using a 3D Radial Phase Encoding (RPE) trajectory. The authors employ the Alternating Direction Method of Multipliers (ADMM) to solve the resulting non-grid optimization problem iteratively.

Results and Discussion

The method was evaluated using in-vivo 3D whole-heart MR data from five healthy volunteers, retrospectively undersampled at factors of $R=8, 16,$ and $24$ (equivalent to scan times as low as 1.6 minutes).

Visual Fidelity and 3D Continuity

A key insight of this work is the power of 3D spatial continuity. By utilizing information across the entire 3D volume, 3DShearCS better distinguishes between incoherent undersampling artifacts and actual anatomical structures compared to 2D approaches.

Visual comparison of 3DShearCS vs 2DShearCS across various undersampling factors (R). Zoomed-in regions highlight the superior vessel depiction of the 3D approach.

Quantitative Performance

3DShearCS consistently demonstrated lower relative errors and higher structural similarity (HaarPSI) than standard itSENSE, Total Variation (TV), and wavelet-based methods. At the highest undersampling factor ($R=24$), 3DShearCS improved coronary artery sharpness by approximately 40% compared to itSENSE and WaveCS.

Quantitative metrics comparing 2DShearCS and 3DShearCS. The 3D approach shows higher vessel sharpness, higher structural similarity (HaarPSI), and lower relative error across all acceleration factors.

Two blinded clinical experts scored 3DShearCS as having the highest image quality and best agreement with reference images, particularly at high acceleration factors ($R=16$ and $R=24$).

Conclusion

3DShearCS significantly accelerates 3D CMR, potentially reducing scan times from over 15 minutes to under 2 minutes while maintaining high image quality. While current reconstruction times (approx. 4 hours on a CPU) remain a challenge, future work focusing on GPU acceleration could make this mathematically rigorous approach a practical clinical tool. The study confirms that leveraging the mathematical properties of shearlets to represent curved anatomical structures is a superior strategy for high-resolution cardiac imaging.

Summary updated

Abstract

High-resolution three-dimensional (3D) cardiovascular magnetic resonance (CMR) is a valuable medical imaging technique, but its widespread application in clinical practice is hampered by long acquisition times. Here we present a novel compressed sensing (CS) reconstruction approach using shearlets as a sparsifying transform allowing for fast 3D CMR (3DShearCS). Shearlets are mathematically optimal for a simplified model of natural images and have been proven to be more efficient than classical systems such as wavelets. Data is acquired with a 3D Radial Phase Encoding (RPE) trajectory and an iterative reweighting scheme is used during image reconstruction to ensure fast convergence and high image quality. In our in-vivo cardiac MRI experiments we show that the proposed method 3DShearCS has lower relative errors and higher structural similarity compared to the other reconstruction techniques especially for high undersampling factors, i.e. short scan times. In this paper, we further show that 3DShearCS provides improved depiction of cardiac anatomy (measured by assessing the sharpness of coronary arteries) and two clinical experts qualitatively analyzed the image quality.
BibTeX
@misc{ma2017shearlet,
  author = {Jackie Ma and Maximilian März and Stephanie Funk and Jeanette Schulz-Menger and Gitta Kutyniok and Tobias Schaeffter and Christoph Kolbitsch},
  title = {{Shearlet-based compressed sensing for fast 3D cardiac MR imaging using iterative reweighting}},
  year = {2017},
  eprint = {1705.00463},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1705.00463},
}
Figure