Categories
Resources

Learned MR Pulse Sequences for T2 and T1ρ Mapping

MATLAB scripts for learned pulse sequence parameters of magnetization-prepared gradient echo sequences used in multi-component T2 and T1ρ mapping.

Share on LinkedIn

We are sharing MATLAB scripts for L-MPGRE: learned magnetization-prepared gradient echo (MPGRE) sequences used in T2 and T1ρ mapping. This two-level machine learning approach can automatically find parameters of MR pulse sequences to achieve desired performance in key criteria such as accuracy, precision, and signal-to-noise ratio (SNR).

Diagram of a two-level machine learning approach.
A diagram of a of a two-level learning approach with differentiable and non-differentiable optimization loops used for L-MPGRE sequences. In the inner loop (differentiable optimization), the framework quickly learns a new pulse sequence by evaluating a surrogate loss function on the predicted signal evolutions. In the outer loop (non-differentiable optimization), the framework performs a more careful evaluation using simulated image reconstruction and multi-objective loss function.

The scripts improve upon our previous methods with two major enhancements:

  • Ability to automatically learn discrete and continuous parameters of the MR pulse sequence.
  • User options to specify performance goals, such as SNR and quantitative accuracy, through a multi-objective loss function.

In a recently submitted manuscript currently under review, we compare this approach to MR sequences commonly used for mono-exponential, stretched-exponential, and bi-exponential T2 and T1ρ mapping in knee cartilage. One example of such common sequence is magnetization-prepared angle-modulated partitioned k-space spoiled GRE snapshots (MAPSS). In the manuscript, we find that when compared to sequences like MAPSS, our L-MPGRE approach improves the quality of the data while reducing scan time by half.

A panel comparing quantitative maps of knee cartilage obtained with MAPSS and L-MPGRE techniques.
A comparison of quantitative T1ρ maps (left, outlined in blue) and T2 maps (right, outlined in red) of knee cartilage obtained in real acquisitions. For each contrast, we show maps obtained with the MAPSS sequence (left) and those obtained with an L-MPGRE sequence (right). The MAPSS sequence takes 20 minutes of scan time and achieves an SNR of 9.6, while the learned sequence, L-MPGRE, takes 10 minutes and achieves an SNR of 15.1, yielding higher quality in half the time.

Get the Code

The software available on this page is provided free of charge and comes without any warranty. CAI²R and NYU Grossman School of Medicine do not take any liability for problems or damage of any kind resulting from the use of the files provided. Operation of the software is solely at the user’s own risk. The software developments provided are not medical products and must not be used for making diagnostic decisions.

The software is provided for non-commercial, academic use only. Usage or distribution of the software for commercial purpose is prohibited. All rights belong to the author (Marcelo Wust Zibetti) and NYU Grossman School of Medicine. If you use the software for academic work, please give credit to the author in publications and cite the related publications.

Please spell out your affiliation (e.g. “New York University” rather than “NYU”).

Contact

Questions about this resource may be directed to Marcelo V. Wust Zibetti, PhD.

Related Story

Related Resources