The Pulseq-CEST Library: definition of preparations and simulations, example data, and example evaluations

Alexander Y. Liebeskind, Jan Rüdiger Schüre, Moritz Fabian, Simon Weinmüller, Patrick Schünke, Vladimir Golkov, Daniel Cremers, Moritz ZaißMagnetic 2025 · DOI

TL;DR

The Pulseq-CEST Library is an open-source repository designed to standardize Chemical Exchange Saturation Transfer (CEST) MRI. By providing a unified framework for pulse sequence definition, Bloch–McConnell simulations, and evaluation scripts, it enables researchers to prototype new protocols in silico and ensures reproducible results across different scanner vendors.

What Problem Does This Paper Address?

CEST MRI is a powerful technique that detects low-concentration metabolites (like proteins or glucose) by measuring their proton exchange with water. However, the field suffers from a "wild west" problem: every research group uses different radiofrequency (RF) pulse shapes, saturation times, and post-processing methods.

Because the CEST signal is incredibly sensitive to these hardware and software parameters, comparing results between two different studies is often like comparing apples to oranges. This lack of standardization hinders clinical adoption and makes it difficult to generate reliable "ground truth" data for emerging deep-learning applications in MRI.

Methods

The authors leverage Pulseq, an open-source framework that allows pulse sequences to be programmed in MATLAB or Python and exported as vendor-neutral text files. The Pulseq-CEST Library organizes the workflow into four modular components:

  1. Preparation Sequences: Standardized .seq files (e.g., APT-weighted or WASABI protocols).
  2. Bloch–McConnell Parameters: YAML files defining the physical properties of the "environment" (exchange rates, T1/T2 relaxation times, and pool concentrations).
  3. Simulation Engine: A validated Bloch–McConnell solver that uses the Padé algorithm to compute how the magnetization evolves during the sequence.
  4. Evaluation Scripts: Standardized Python/MATLAB scripts for calculating Z-spectra and MTRasym (magnetization transfer ratio asymmetry) maps.

This modularity allows a researcher to hold the "environment" constant while testing a dozen different pulse sequences, or vice versa, all without needing physical scanner time.

Datasets

The paper validates the library using both simulated and physical data:

  • The Five-Tube Phantom: A physical phantom containing five 50 mL tubes with varying concentrations of L-arginine (20–70 mM), agarose (to mimic tissue-like semi-solid components), and Gadovist (to adjust T1 relaxation).
  • Simulated Environments: Digital "phantoms" for L-arginine, creatine, and white matter (the latter featuring four distinct CEST pools and one Nuclear Overhauser Effect (NOE) pool).
  • Benchmark Data: The authors provide a specific dataset (accessible via Zenodo) to allow others to test their own reconstruction pipelines against the paper's results.

Key Insights and Ideas

  • The "Digital Twin" for Artifacts: One of the most interesting findings was that the simulation successfully reproduced "sidebands"—artifacts usually seen in liquid phantoms. This allows researchers to distinguish between actual metabolic signals and hardware-induced artifacts.
  • Interpolation Matters: The authors demonstrated that the choice of interpolation (linear vs. smoothing spline) in post-processing significantly impacts the final Z-spectrum. Splines make the data look "cleaner" but can accidentally smooth over real, subtle metabolic peaks.
  • Quantifying Divergence: By simulating white matter using parameters from four different landmark studies, the authors showed significant divergence in the resulting spectra. This highlights exactly why a library like this is necessary: even when studying the same tissue, different assumptions lead to different results.

Key Results

  • Protocol Validation: The library successfully matched real-world phantom data with simulations for both APTw (Amide Proton Transfer weighted) and WASABI (Water shift and B1) protocols.
  • Peak Alignment: MTRasym plots for L-arginine consistently peaked at +3 ppm in both real and simulated data, confirming the accuracy of the simulation's exchange models.
  • Cross-Vendor Portability: Because the sequences are defined in the Pulseq format, they were shown to be transferable across different scanner manufacturers, providing a path toward hardware-independent MRI protocols.

Limitations and Future Work

The primary limitation is that the library’s utility depends on community adoption; its "standardizing" power only works if the community contributes new sequences and environment definitions.

While the current work focuses heavily on the CEST effect, the authors note that the framework is naturally suited for expansion into other saturation transfer techniques, such as relayed NOEs (rNOEs) and semisolid magnetization transfer contrast (MTC). Future iterations may also integrate more deeply with image processing toolkits like MITK to bridge the gap between raw simulation and clinical image segmentation.

Summary updated

Abstract

The Pulseq-CEST Library provides a flexible mechanism for standardizing and prototyping CEST sequences, facilitating collaborative development. With the capability for expansion, including open-source incorporation of new sequences and environments, the library accelerates the invention and spread of novel CEST and other saturation transfer approaches, such as relayed NOEs (rNOEs) and semisolid magnetization transfer contrast (MTC) methods.
BibTeX
@inproceedings{liebeskind2025pulseq,
  author = {Alexander Y. Liebeskind and Jan Rüdiger Schüre and Moritz Fabian and Simon Weinmüller and Patrick Schünke and Vladimir Golkov and Daniel Cremers and Moritz Zaiß},
  title = {{The Pulseq-CEST Library: definition of preparations and simulations, example data, and example evaluations}},
  booktitle = {Magnetic},
  year = {2025},
  doi = {10.1007/s10334-025-01242-6},
}
Figure