Enhancing SNR in CEST imaging: A deep learning approach with a denoising convolutional autoencoder.

Area of Science:

• Magnetic Resonance Imaging
• Biomedical Engineering
• Artificial Intelligence

Background:

• Chemical Exchange Saturation Transfer (CEST) imaging is a valuable technique for detecting molecular changes in tissues.
• However, CEST images often suffer from low signal-to-noise ratio (SNR), limiting their diagnostic potential.
• Developing effective denoising methods is crucial for improving CEST imaging quality.

Purpose of the Study:

• To develop a novel denoising convolutional autoencoder (DCAE) for SNR enhancement in CEST imaging.
• To compare the performance of the proposed DCAE-CEST method against state-of-the-art denoising techniques.

Main Methods:

• A DCAE-CEST model was developed, comprising an encoder and decoder network utilizing 1D convolutions and pooling/up-sampling layers.
• The model was trained using simulated CEST Z-spectra with Kullback-Leibler divergence and Principal Component Analysis (PCA) for context learning.
• Performance was evaluated using simulated data and in vivo data from an animal tumor model, quantifying amide proton transfer (APT) and nuclear Overhauser enhancement (NOE) maps.

Main Results:

• The DCAE-CEST method demonstrated superior performance in digital phantom experiments, outperforming existing denoising techniques in peak SNR and Structural Similarity Index.
• In vivo data confirmed the effectiveness of DCAE-CEST in denoising APT and NOE maps.
• Significant differences in NOE, but not APT, were observed between tumor and normal tissues in vivo.

Conclusions:

• The DCAE-CEST model effectively learns key features of the CEST Z-spectrum.
• It provides a superior denoising solution compared to existing methods for CEST imaging.
• This approach holds promise for improving the diagnostic utility of CEST imaging.