FlexCENT: A frequency-flexible CEST imaging network combining frequency offset encoding and three-dimensional U-Net.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biophysics

Background:

• Chemical Exchange Saturation Transfer (CEST) imaging is a powerful MRI technique for detecting metabolites.
• Current CEST quantification methods often require specific frequency offset schemes and retraining for new protocols.
• Robust and flexible CEST quantification is crucial for broader clinical adoption.

Purpose of the Study:

• To develop a deep learning-based method, FlexCENT, for robust and frequency-flexible CEST quantification.
• To enable accurate CEST parameter estimation across variable frequency offset schemes without retraining.

Main Methods:

• FlexCENT integrates frequency offset encoding with a 3D U-Net architecture.
• It processes CEST images and frequency offsets to predict Lorentzian parameters of a 4-pool model, including B0 inhomogeneity.
• Frequency offsets are transformed into a continuous spectral feature representation for generalization.

Main Results:

• FlexCENT demonstrated successful CEST quantification across numerical simulations, preclinical (mouse tumor), and clinical (human brain) studies.
• The network maintained consistent performance under varying frequency offset conditions without retraining.
• FlexCENT showed superior noise robustness and enhanced anatomical delineation in vivo parametric mapping compared to existing methods.

Conclusions:

• FlexCENT provides an efficient, flexible, and robust quantitative approach for CEST imaging by combining spectral and spatial information.
• This method significantly enhances the quantification capability and clinical potential of CEST imaging.
• FlexCENT overcomes limitations of previous methods by generalizing to unseen frequency offset schemes.