Denoising Low-Power CEST Imaging Using a Deep Learning Approach With a Dual-Power Feature Preparation Strategy

Area of Science:

• Magnetic Resonance Imaging
• Biomedical Engineering
• Artificial Intelligence

Background:

• Low-power (LP) chemical exchange saturation transfer (CEST) Z-spectra offer improved peak resolvability but suffer from low contrast-to-noise ratio (CNR).
• Denoising LP Z-spectra is crucial for accurate observation and quantification of CEST effects in various applications.

Purpose of the Study:

• To develop a dual-power feature preparation for an autoencoder-based deep learning approach (DPDL) to denoise LP Z-spectra.
• To leverage the high CNR of higher saturation power and the enhanced peak resolvability of low saturation power for improved CEST imaging.

Main Methods:

• The DPDL model was trained on simulated CEST data and validated on phantoms and in vivo rat brain and leg muscles at 4.7T.
• Lorentzian difference (LD) analysis quantified CEST effects, and peak signal-to-noise ratio (PSNR) assessed denoising performance.
• DPDL was compared against existing denoising methods using equivalent acquisition times.

Main Results:

• DPDL demonstrated superior PSNR in phantom experiments compared to existing techniques.
• In vivo experiments showed improved image quality and revealed key tissue component peaks in rat brain and muscle.
• The method outperformed existing denoising techniques in animal studies.

Conclusions:

• The DPDL method offers superior denoising for LP CEST imaging, enhancing the isolation of various chemical pools.
• This advancement improves CEST applications, particularly in low-field MRI settings.
• DPDL effectively addresses the CNR limitations of LP Z-spectra.