Magnetic resonance imaging (MRI) has well-established applications spanning a wide array of practice areas. Chemical exchange saturation transfer (CEST) MRI describes an advanced MRI technique that relies on detecting proton exchange between water and solute molecules in biological tissues. This allows CEST to yield imaging that highlights targeted molecular components and metabolites, granting visibility into tissue properties that can help elucidate brain tumors and other abnormalities.

Given the importance of optimizing MRI approaches, a significant body of research has been dedicated to understanding pulse sequences, Bloch–McConnell simulations, and evaluation methods. Since it can often be more straightforward to work with pulse sequences prior to deployment within a physical MR machine, open-source frameworks have emerged that allow for the development of pulse sequence parameters in silico. Pulseq, for example, supports pulse sequence programming in MATLAB and Python, generating text files that work in simulation but also in MR scanners [1]. While image reconstruction functions are commonly vendor-specific and proprietary, it is possible to use tailored Pulseq interpreters with Pulseq pulse sequences to provide compatibility with established MRI readouts [e.g., fast low angle shot (FLASH), gradient-recalled echo (GRE), echo planar imaging (EPI)] [2]. In the context of CEST, Pulseq is, therefore, ideal because it delivers portability, transparency, and therefore reproducibility: CEST preparation periods are accessible and interpretable, easy to work with in prevalent programming languages, directly transferable to MATLAB or Python simulations, and readily functional in real scanners.

As the usage of different parameter sets for CEST imaging—including specifications such as irradiation strength (B1), total saturation time (\(t_\)), saturation duty cycle (DC), radiofrequency (RF), pulse shape (magnitude, phase), pulse duration (\(t_p\)), pulse delay (\(t_d\)), and phase cycling (\(\phi _i\))—has not been standardized so far, an important step for quantitative and comparable CEST imaging is standardization to prevent the emergence of different signals [3, 4]. To facilitate this task, Pulseq-CEST [2] was launched to create a consensus on preparation methods and encourage sharing of CEST parameters, though efforts to standardize CEST methodology remain a work in progress [5, 6]. As a result of the Pulseq-CEST open standard, researchers have been able to share pre-saturation schemes in a recognized format in various studies [3, 7, 8]. Beyond enhancing reproducibility, Pulseq-CEST data have also been used in deep learning as ground truth [9]. This application is increasingly important given the growing usage of deep learning in MR sequencing and the essential nature of a baseline consensus as a means of evaluating model performance.

Achieving standardization with Pulseq-CEST requires accounting for multiple components. Starting from the pulse sequences, it is essential to set up established protocols including the specified CEST sequences (e.g., amide proton transfer weighted (APTw) [10]), which can optionally be followed by a water shift and B1 (WASABI) sequence to perform B0 and B1 correction in post-processing [11]. These protocols must then be applied to consistent environments, including real and simulated data. In other to maintain cross-study comparability and reproducibility, the evaluation scripts used must also involve the same steps.

The Pulseq-CEST Library, building upon the introduction of the initial Pulse-CEST standard to drive towards a more robust understanding of CEST, provides sequences, simulation environments, phantoms, and evaluation methods to streamline working with CEST across protocols. These examples provide a common basis for working with popular pulse sequences and simulation environments, laying the foundation for more universal and reproducible MRI research across CEST and other saturation transfer approaches including relayed NOEs (rNOEs) and semisolid magnetization transfer contrast (MTC) methods [12]. This work focuses on providing demonstrations with the CEST effect.

Using the Pulseq-CEST Library, one candidate sequence or environment can be thoroughly tested while changing the other input parameters. This allows for the various components of a Pulseq-CEST simulation to be standardized, or for new sequences and environments to be developed and shared. For example, if an L-arginine environment is the object for CEST investigations, the simulation environment can be held constant while varying between different sequence parameters. Other sequences can also be deployed to further investigate the simulation environment in the context of protocols for glutamate [13], pH-weighted [14, 15], WASABI [16], or dynamic glucose enhanced (DGE) [17] at different field strengths such as 3T and 7T. This allows for comparison of entire protocols and discrete evaluation of individual changes—a user could determine how changing the number of pulses or flip angle directly changes the resulting spectra and parameter maps for instance. When designing new contrasts, this ability is critical, as novel pulse sequences can be designed and tested in silico without necessitating a real scanner. Due to the format of Pulseq files, a resulting sequence can then be used in scanners across vendors, allowing for confirmation of simulated results in phantoms and in vivo.

A major benefit of this approach is therefore cross-vendor compatibility. Beyond the pragmatic considerations mentioned above, developing and experimenting with pulse sequences prior to interpretation provides portability across manufacturers, allowing for a hardware independent approach. The experimental data can then be used to further iterate upon Pulseq sequence files. In practice, the Pulseq-CEST library can be divided into three sections as demonstrated in Fig. 1. The full library is accessible to the public through the repository at https://github.com/kherz/pulseq-cest-library. while the code demonstrations used here are located https://github.com/kherz/pulseq-cest-library/releases/tag/MAGMA_2025_PAPER. The associated phantom data can be found at https://zenodo.org/records/15053086.

Illustration of the Pulseq-CEST pipeline, which is based on Pulseq. Standardized preparations, simulations, and evaluations provide the basis for reproducible experiments that can be replicated across simulated environments and vendor-specific hardware using tailored interpreters [18]

The library can be utilized to simulate distinct chemical exchange mechanisms given the differing use cases for MRI protocols. APT-weighted imaging, for instance, is commonly used to identify tumors exhibiting unusual pH levels [15], elevated protein contents [19], or increased cellularity in high-grade gliomas. In this type of imaging, a series of RF pulses is applied in a symmetrical manner offset from water resonance with additional saturation parameters [3], provided in the seq-files of the seq-library. These parameters can be adjusted for each specific task or for standardization. In addition to the APT-weighted protocols or other examples like glucose-weighted CEST (glucoCEST), B0 and B1 correction can be performed according to the WASABI sequence [11, 20]. Furthermore, the sim-library offers various environments in which the respective seq-files can be simulated, while the eval examples provide a simple evaluation of the respective sequences.

In short, the Pulseq-CEST Library makes it easier not only to measure, but also to simulate and experiment with imaging techniques that can be used to target specialized physiology. In the following work, we concentrate on the evaluation of CEST data.