Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Transformer-based Parameter Fitting of Models derived from Bloch-McConnell Equations for CEST MRI Analysis

Christof Duhme, Chris Lippe, Verena Hoerr, Xiaoyi Jiang

TL;DR

Quantifying CEST MRI signals is challenging due to the interaction of many physiological and experimental parameters described by Bloch–McConnell dynamics. The authors present a transformer‑based encoder–decoder network trained in a self‑supervised manner to fit parameters of Lorentzian, analytical Z, and MTR_Rex models to in‑vitro CEST spectra, enforcing physical bounds through a model‑aware loss. Across phantom data, the approach outperforms traditional gradient‑based solvers in accuracy and consistency while delivering substantial runtime speedups on GPU. This work provides a robust, fast parameter estimation framework for CEST MRI and establishes a foundation for extending to in vivo data with spatial regularization.

Abstract

Chemical exchange saturation transfer (CEST) MRI is a non-invasive imaging modality for detecting metabolites. It offers higher resolution and sensitivity compared to conventional magnetic resonance spectroscopy (MRS). However, quantification of CEST data is challenging because the measured signal results from a complex interplay of many physiological variables. Here, we introduce a transformer-based neural network to fit parameters such as metabolite concentrations, exchange and relaxation rates of a physical model derived from Bloch-McConnell equations to in-vitro CEST spectra. We show that our self-supervised trained neural network clearly outperforms the solution of classical gradient-based solver.

Transformer-based Parameter Fitting of Models derived from Bloch-McConnell Equations for CEST MRI Analysis

TL;DR

Quantifying CEST MRI signals is challenging due to the interaction of many physiological and experimental parameters described by Bloch–McConnell dynamics. The authors present a transformer‑based encoder–decoder network trained in a self‑supervised manner to fit parameters of Lorentzian, analytical Z, and MTR_Rex models to in‑vitro CEST spectra, enforcing physical bounds through a model‑aware loss. Across phantom data, the approach outperforms traditional gradient‑based solvers in accuracy and consistency while delivering substantial runtime speedups on GPU. This work provides a robust, fast parameter estimation framework for CEST MRI and establishes a foundation for extending to in vivo data with spatial regularization.

Abstract

Chemical exchange saturation transfer (CEST) MRI is a non-invasive imaging modality for detecting metabolites. It offers higher resolution and sensitivity compared to conventional magnetic resonance spectroscopy (MRS). However, quantification of CEST data is challenging because the measured signal results from a complex interplay of many physiological variables. Here, we introduce a transformer-based neural network to fit parameters such as metabolite concentrations, exchange and relaxation rates of a physical model derived from Bloch-McConnell equations to in-vitro CEST spectra. We show that our self-supervised trained neural network clearly outperforms the solution of classical gradient-based solver.
Paper Structure (14 sections, 11 equations, 1 figure, 2 tables)

This paper contains 14 sections, 11 equations, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Physical Background of CEST MRI
  4. CEST metrics
  5. CEST models
  6. Fitting the Physical Models
  7. Transformer-based Neural Network
  8. Performance analysis
  9. Experiments & Results
  10. Dataset
  11. Experimental Setup
  12. Results & Discussion
  13. Runtime
  14. Conclusion

Figures (1)

  • Figure 1: Example plots of proton fractions for lactate for the $MTR_{R_{ex}}$ model. Dashed lines are the parameters of the model with the colors indicating the concentration of glucose, bold lines are the result of the linear regression model, errorbars are computed over the test set.