Variable density sampling based on physically plausible gradient waveform. Application to 3D mriangiography

Nicolas Chauffert, Pierre Weiss, Marianne Boucher, Sébastien Mériaux, Philippe CiuciliISBI 2015 · DOI

TL;DR

This paper introduces a novel method for designing MRI gradient waveforms that enables faster and more efficient k-space sampling, particularly for variable density trajectories. By projecting desired sampling curves onto the physical constraints of MRI hardware, they achieve significant reductions in scan time while maintaining or improving image reconstruction quality, demonstrated on 3D MRI angiography.

What Problem Does This Paper Address?

Magnetic Resonance Imaging (MRI) often relies on acquiring data in k-space, the spatial frequency domain. To speed up scans, researchers use "variable density sampling," where more data points are collected in important regions of k-space and fewer in less critical ones. This is theoretically beneficial for image reconstruction, especially with Compressed Sensing (CS) techniques. However, a major hurdle is translating these desired sampling patterns into actual MRI scans. Existing methods for designing the magnetic field gradients needed to traverse these k-space paths can be very slow, especially when the desired trajectory has sharp turns or high curvature. This slowness negates the time savings gained from variable density sampling. This paper tackles the problem of designing physically plausible gradient waveforms that can efficiently execute these variable density trajectories, overcoming the limitations of existing methods.

Methods

The core of this paper's contribution lies in a novel projection algorithm for gradient waveform design. Instead of trying to perfectly follow a desired k-space trajectory with optimal control (which can be slow), their method takes an input trajectory and projects it onto the set of physically admissible trajectories that respect the hardware constraints of the MRI scanner (maximum gradient strength and slew rate).

Here's a breakdown:

  • Input: A desired k-space trajectory, often generated by methods like the Traveling Salesman Problem (TSP) to approximate a variable density function (π-VDS).
  • Projection Algorithm: This algorithm finds the closest trajectory in the set of "physically plausible" curves. This means the resulting trajectory respects the scanner's gradient limits (magnitude and slew rate). The mathematical formulation involves minimizing the difference between the input curve and an admissible curve, solved using iterative proximal gradient methods.
  • Key Advantage: This projection approach allows for faster traversal of the k-space trajectory compared to traditional optimal control methods, especially for trajectories with high curvature. It also fixes the acquisition time to that of the input curve, often resulting in a shorter overall scan time.

Datasets

The paper uses simulated data for initial proof-of-concept and then evaluates their method on experimental data from a 17.2T preclinical MRI scanner. The experimental data involves 3D MRI angiography of a mouse brain, both before and after the injection of a contrast agent. No new benchmarks are introduced; rather, the method is evaluated against existing approaches using standard metrics.

Key Insights and Ideas

  • Projection is Key: The central idea that projecting a desired, potentially unrealistic, k-space trajectory onto the set of physically realizable ones is a powerful way to design efficient gradient waveforms.
  • TSP + Projection = Efficiency: Combining TSP-based variable density sampling with their projection algorithm offers a practical way to implement complex sampling schemes that are both efficient and physically achievable.
  • Time vs. Quality Trade-off: The method allows for a more flexible trade-off between acquisition time and image reconstruction quality. One can either fix the scanning time and get better reconstructions, or achieve a given reconstruction quality in less time.
  • Overcoming Curvature Bottlenecks: The projection method effectively handles high-curvature regions in trajectories, which are major bottlenecks for traditional optimal control approaches.

Key Results

  • Faster Traversal: The projection algorithm significantly reduces the time required to traverse k-space trajectories compared to optimal control methods, especially for TSP-based variable density sampling.
  • Improved Reconstruction Quality: For a fixed number of samples, their method yields comparable or better image reconstruction quality.
  • Significant Time Reduction in 3D Angiography: In 3D MRI angiography of mouse brains, their method achieved a 75% reduction in acquisition time (from 3.53s to 0.88s) with an acceleration factor of 17.4, while still recovering fine vascular structures.
  • Flexibility: The method can implement variable density trajectories that approximate any desired density function, offering more control over the sampling process.

Limitations and Future Work

  • Real-World Scanner Implementation: The authors mention they are currently implementing these waveforms on actual scanners to validate their method in a real-world CS-MRI framework. This suggests that the current results are primarily based on simulations and retrospective reconstructions.
  • Approximation of k-space Grid: The reconstruction process involves projecting onto a discrete k-space grid, which can introduce approximations. The authors note that non-Cartesian reconstruction methods could potentially avoid this.
  • Future Work: The immediate future work involves validating the method on actual scanners. Further exploration could involve more complex hardware constraints (rotation-variant) and integration with advanced reconstruction techniques.

Summary updated

BibTeX
@inproceedings{chauffert2015variable,
  author = {Nicolas Chauffert and Pierre Weiss and Marianne Boucher and Sébastien Mériaux and Philippe Ciucili},
  title = {{Variable density sampling based on physically plausible gradient waveform. Application to 3D mriangiography}},
  booktitle = {ISBI},
  year = {2015},
  doi = {10.1109/isbi.2015.7164154},
}
Figure