Synthesis of higher-B₀ CEST Z-spectra from lower-B₀ data via deep learning and singular value decomposition

Area of Science:

• Magnetic Resonance Imaging
• Biomedical Engineering
• Artificial Intelligence

Background:

• Chemical Exchange Saturation Transfer (CEST) MRI at 3 Tesla (3T) faces challenges with metabolite specificity due to overlapping signals.
• Higher magnetic field strengths (B0) offer better Z-spectral peak separation for improved CEST MRI interpretation and quantification.
• Access to high-field MRI scanners is limited by equipment availability, field homogeneity, and safety concerns.

Purpose of the Study:

• To develop and validate a deep learning framework for synthesizing high-field (e.g., 9.4T) Z-spectra from lower-field (3T) CEST MRI data.
• To overcome the limitations of lower field strengths in metabolite signal separation and quantification.
• To enable more accessible and robust CEST MRI applications by leveraging existing lower-field infrastructure.

Main Methods:

• A deep learning framework combining two deep neural networks (DNNs) and singular value decomposition (SVD) was trained using Bloch-McConnell equation simulations.
• The first DNN corrects for B0 shifts and aligns Z-spectra frequencies.
• The second DNN transforms lower-field spectra into higher-field representations via SVD truncation, followed by inverse SVD reconstruction.

Main Results:

• The synthesized 9.4T Z-spectra closely matched experimental ground truth across phantoms and in vivo rat brains (low RMSE: 0.11%-1.8%).
• High R-squared values (>0.99) and accurate synthesized contrast maps (amide and NOE) were achieved.
• The framework demonstrated robustness against B0 inhomogeneities, noise, and acquisition imperfections.

Conclusions:

• The proposed deep learning framework successfully synthesizes higher-B0 Z-spectra from lower-B0 CEST MRI data.
• This approach enhances spectral resolution and quantification capabilities without requiring higher-field scanners.
• The method holds significant potential for advancing CEST MRI applications in clinical and research settings.