High-Quality CEST Mapping With Lorentzian-Model Informed Neural Representation.

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Biophysics
• Medical Diagnostics

Background:

• Chemical Exchange Saturation Transfer (CEST) MRI enhances detection of low-concentration macromolecules and metabolites.
• Conventional CEST mapping methods (model-based and deep learning) have limitations in sensitivity, robustness, and generalizability.
• Accurate CEST mapping is crucial for quantifying molecular information in various biological contexts.

Purpose of the Study:

• To develop a novel framework, Lorentzian-model Informed Neural Representation (LINR), for high-quality CEST mapping.
• To overcome the limitations of existing model-based and data-driven CEST quantification methods.
• To enable sensitive and generalizable molecular detection using MRI.

Main Methods:

• Proposed a self-supervised neural architecture, LINR, embedding the Lorentzian equation for CEST signal modeling.
• Directly reconstructed high-sensitivity parameter maps from raw z-spectra, eliminating the need for labeled training data.
• Theoretically guaranteed convergence of the self-supervised training strategy for mathematical validity.

Main Results:

• LINR demonstrated superior performance in capturing CEST contrasts compared to conventional methods.
• Evaluations on synthetic phantoms and in-vivo experiments (tumor, Alzheimer's models) confirmed LINR's effectiveness.
• The framework exhibited high sensitivity and robustness in CEST mapping.

Conclusions:

• LINR provides a parameter-free and versatile tool for advanced CEST mapping.
• The framework facilitates non-invasive molecular diagnostics and pathophysiological discovery.
• LINR's adaptive integration into diverse MRI workflows enhances its clinical potential.