Accelerated and quantitative three-dimensional molecular MRI using a generative adversarial network.

Area of Science:

• Magnetic Resonance Imaging
• Quantitative Imaging
• Biophysical Modeling

Background:

• Quantitative 3D chemical exchange saturation transfer (CEST) and semisolid magnetization transfer (MT) imaging are crucial for various medical applications.
• Current acquisition times for these advanced MRI techniques can be prohibitively long, limiting clinical utility.
• Rapid reconstruction of quantitative exchange parameter maps is essential for efficient workflow.

Purpose of the Study:

• To substantially shorten the acquisition time for quantitative 3D CEST and MT imaging.
• To enable rapid reconstruction of chemical exchange parameter maps.
• To develop a robust method applicable across different scanner models and patient populations.

Main Methods:

• A novel saturation transfer-oriented generative adversarial network (GAN-ST) was developed.
• GAN-ST was trained on 3D CEST and MT magnetic resonance fingerprinting (MRF) datasets from phantoms and diverse patient groups (healthy, cancer, cardiac).
• The framework learns a mapping from reduced data to quantitative exchange parameter space.

Main Results:

• GAN-ST achieved a 70% reduction in 3D acquisition time (42-52 s) compared to conventional CEST-MRF.
• Quantitative reconstruction of whole-brain parameter maps was achieved in just 0.8 seconds.
• Excellent agreement was observed between GAN-ST results and ground truth for L-arginine concentration and pH (Pearson's r > 0.95).
• High image quality (SSIM > 94%) and accuracy (NRMSE < 7%) were maintained for semi-solid exchange parameters in complex cases, including susceptibility artifacts.

Conclusions:

• GAN-ST significantly reduces acquisition time for quantitative semi-solid MT/CEST mapping.
• The method demonstrates robust performance across different pathologies and scanner variations not seen during training.
• This advancement facilitates faster and more efficient quantitative MRI.