Targeted vessel reconstruction in non‐contrast‐enhanced steady‐state free precession angiography

Efe Ilıcak, Süheyla Çetin, Elif Bulut, Kader Karlı Oğuz, Emine Ülkü Sarıtaş, Gözde Ünal, Tolga ÇukurNMR 2016 · DOI

TL;DR

This paper introduces a novel "Targeted Compressed Sensing" (TCS) method for non-contrast-enhanced Magnetic Resonance Angiography (NCE MRA). By using automatically generated 3D vessel maps to adaptively apply sparsity penalties, TCS significantly improves blood-to-background contrast and vessel depiction, overcoming limitations of standard compressed sensing techniques.

What Problem Does This Paper Address?

Non-contrast-enhanced MRA (NCE MRA) is a valuable tool for visualizing blood vessels without the risks associated with contrast agents. However, achieving high-quality NCE MRA images is challenging due to the need for rapid image acquisition, which often leads to undersampling of the data. This undersampling introduces aliasing artifacts and noise, degrading image quality, particularly the contrast between blood and surrounding tissues. Standard compressed sensing (CS) techniques, which aim to reconstruct images from undersampled data using sparsity priors, typically apply uniform penalties. This uniform weighting can either fail to adequately suppress background noise and aliasing, or it can suppress important low-contrast vessel signals, creating a trade-off between contrast and vessel preservation. The paper addresses this by proposing a method to improve both blood-background contrast and vessel depiction in NCE MRA, especially at accelerated scan rates.

Methods

The core of the proposed method, Targeted Compressed Sensing (TCS), is a two-stage reconstruction process:

  1. Tractographic Segmentation for Vasculature Mapping: The first stage involves generating a 3D map of the vasculature. This is achieved by first performing a standard zero-filled Fourier reconstruction of the undersampled data. Then, a sophisticated segmentation algorithm, which jointly models tubular sections and branching structures using a fourth-order tensor model, is applied to extract the vessel tree from these initial reconstructions. This algorithm can trace entire vessel trees in under three minutes.
  2. Spatially Adaptive Penalties: The generated vasculature map is then used to guide the reconstruction in the second stage. Instead of applying uniform sparsity and total variation (TV) penalties, TCS applies spatially adaptive penalties. Specifically, the sparsity (l₁-norm) penalty is made heavier in background regions to suppress noise and aliasing, while it's reduced within the vessels to preserve their signals. Conversely, the TV penalty, which promotes image homogeneity, is reduced in vessel regions to avoid blurring fine details and is increased in the background for denoising. This adaptive weighting is achieved by creating a "penalty-weight map" derived from the segmented vasculature, which is then used to scale the penalty terms.

The optimization problem is formulated as: $$ \min_m ||Fum - Y||_2^2 + ||\lambda_1 \circ W \circ m||_1 + ||\lambda_2 \circ (W/r) \circ \Delta m||_1 $$ where $m$ is the reconstructed image, $Y$ is the acquired data, $F$ is the Fourier transform, $\lambda_1$ and $\lambda_2$ are base penalty weights, $W$ is the spatially adaptive weight map derived from the vasculature segmentation, and $r$ is a parameter controlling the TV penalty scaling.

Datasets

The study utilized both phantom data and in vivo human imaging data for evaluation.

  • Phantom Data: Two numerical phantoms were created to simulate vessels immersed in muscle tissue. These phantoms contained varying numbers and sizes of circular vessels, mimicking peripheral vasculature. The phantoms were used to evaluate the method's performance under controlled conditions, including varying acceleration factors and simulating segmentation errors.
  • In Vivo Data: Peripheral angiograms were acquired from healthy human subjects. Specifically, data were collected from the hand, lower leg, and foot. These datasets were acquired using a 3D magnetization-prepared steady-state free precession (SSFP) pulse sequence. The data were then retrospectively undersampled at various acceleration factors (R = 1, 2, 4, 6, 8) for evaluation.

Key Insights and Ideas

  • Leveraging Structural Information: The core insight is that the inherent structural information of the vasculature, even from undersampled data, can be effectively used to guide the reconstruction process. By identifying vessel locations, we can intelligently apply regularization, rather than using a one-size-fits-all approach.
  • Targeted Regularization for Contrast Enhancement: The key innovation is the adaptive application of sparsity and TV penalties. Making penalties stronger in the background and weaker in vessels directly targets the problem of poor blood-background contrast in NCE MRA.
  • 3D Segmentation is Crucial: Using a 3D tractographic segmentation approach provides a more robust and complete representation of the vasculature compared to 2D methods, leading to more accurate adaptive weighting.
  • Trade-off Relaxation: TCS effectively relaxes the traditional trade-off between image contrast, resolution, and scan efficiency. It allows for higher acceleration factors without sacrificing image quality, and in fact, improves contrast.
  • Robustness to Segmentation Errors: Simulations showed that TCS remains effective even with moderate losses in the segmented vessel maps, indicating a degree of robustness.

Key Results

  • Significant Contrast Improvement: TCS consistently demonstrated substantial improvements in blood/muscle contrast compared to conventional CS methods.
  • Hand: 71.3 ± 28.9% improvement over CSlow.
  • Lower Leg: 30.6 ± 11.3% improvement over CSlow.
  • Foot: 28.1 ± 7.0% improvement over CSlow.
  • Improved Vessel Depiction: Qualitative assessments and quantitative resolution measurements indicated that TCS preserves or improves vessel depiction, especially for smaller vessels, while simultaneously enhancing background suppression.
  • Performance at High Acceleration: The benefits of TCS were observed across various acceleration factors (R=1-8), with significant improvements in contrast even at higher R values where standard CS methods struggle.
  • Robustness: Simulations showed that TCS maintained considerable contrast and vessel size accuracy even with up to 30% loss in segmented vessel volume.
  • Expert Evaluation: Radiologists rated TCS as having superior image contrast, vessel demarcation, and distal-branch visualization compared to other CS methods.

Limitations and Future Work

  • Segmentation Dependence: The performance of TCS is inherently dependent on the quality of the segmented vasculature map. At very high acceleration factors (R≥6), the reduction in segmented volume (due to loss of small vessels in the initial reconstruction) can lead to suboptimal depiction of some small vessels.
  • High R Artifacts: At very high R values, the loss of high-spatial-frequency information in the initial reconstruction can lead to artifacts that might mimic stenoses in minor vessel branches.
  • Synovial Fluid: The presence of synovial fluid, which has a high T2/T1 ratio, can mimic vessel signal and is difficult to suppress fully with standard background penalties.
  • Future Directions:
    • Joint Optimization: Casting the segmentation and reconstruction stages as a single joint optimization problem could further improve robustness.
  • Higher-Order Penalties: Exploring higher-order TV terms could offer better denoising while preserving edge information.
  • Wavelet Domain Sparsity: Applying spatially weighted wavelet penalties could be beneficial for specific applications like coronary imaging.
  • Clinical Applications: The authors suggest future studies to examine pathological cases, including abrupt occlusions and small stenotic vessels, and to extend the method to other angiographic acquisitions.
  • Synovial Fluid Segmentation: Developing specific maps for synovial fluid to apply tailored penalties could improve background suppression in joint imaging.

Summary updated

Abstract

Image quality in non-contrast-enhanced (NCE) angiograms is often limited by scan time constraints. An effective solution is to undersample angiographic acquisitions and to recover vessel images with penalized reconstructions. However, conventional methods leverage penalty terms with uniform spatial weighting, which typically yield insufficient suppression of aliasing interference and suboptimal blood/background contrast. Here we propose a two-stage strategy where a tractographic segmentation is employed to auto-extract vasculature maps from undersampled data. These maps are then used to incur spatially adaptive sparsity penalties on vascular and background regions. In vivo steady-state free precession angiograms were acquired in the hand, lower leg and foot. Compared with regular non-adaptive compressed sensing (CS) reconstructions (CSlow ), the proposed strategy improves blood/background contrast by 71.3 ± 28.9% in the hand (mean ± s.d. across acceleration factors 1-8), 30.6 ± 11.3% in the lower leg and 28.1 ± 7.0% in the foot (signed-rank test, P < 0.05 at each acceleration). The proposed targeted reconstruction can relax trade-offs between image contrast, resolution and scan efficiency without compromising vessel depiction.
BibTeX
@inproceedings{ilcak2016targeted,
  author = {Efe Ilıcak and Süheyla Çetin and Elif Bulut and Kader Karlı Oğuz and Emine Ülkü Sarıtaş and Gözde Ünal and Tolga Çukur},
  title = {{Targeted vessel reconstruction in non‐contrast‐enhanced steady‐state free precession angiography}},
  booktitle = {NMR},
  year = {2016},
  doi = {10.1002/nbm.3497},
}
Figure