Enhancing SNR of CEST MRI Through Noise-to-Noise Deep Learning With K-Space Data Consistency

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Chemical Exchange Saturation Transfer (CEST) MRI is crucial for molecular imaging.
• Acquired CEST images often suffer from low signal-to-noise ratio (SNR), limiting diagnostic accuracy.
• Conventional denoising methods may require clean data or complex simulations.

Purpose of the Study:

• To develop a deep learning-based retrospective denoising method for CEST images.
• To enhance the SNR of acquired CEST images using a novel noise-to-noise (N2N) approach.
• To evaluate the performance of the N2NCEST model on in vivo data.

Main Methods:

• Developed the N2NCEST model using a noise-to-noise (N2N) deep learning strategy with noisy CEST image pairs.
• Incorporated a transformer module to leverage spatiotemporal correlations across frequency offsets.
• Integrated a data consistency layer to preserve k-space information from centric-encoded acquisitions.

Main Results:

• The N2NCEST model successfully enhanced SNR in CEST images from wild-type mouse brains.
• Denoising performance was validated using Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM).
• N2NCEST demonstrated superior image fidelity compared to existing denoising methods, preserving high-frequency details.

Conclusions:

• N2NCEST offers an effective and compatible solution for retrospective CEST image denoising.
• The N2N approach avoids complex parameter tuning and data simulation, enhancing usability.
• This method holds promise for improving the diagnostic potential of CEST MRI.