TagGen: Diffusion-based generative model for cardiac MR tagging super resolution.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Magnetic Resonance (MR) tagging is crucial for assessing myocardial motion.
• Accelerated MR tagging is needed for efficient clinical workflows.
• Super-resolution techniques can improve the quality of low-resolution MR tagging images.

Purpose of the Study:

• To develop a cascaded diffusion-based super-resolution model for low-resolution (LR) MR tagging.
• To integrate this model with parallel imaging for highly accelerated MR tagging.
• To enhance the tag grid quality of LR MR tagging images.

Main Methods:

• Introduced TagGen, a diffusion-based conditional generative model for super-resolution.
• Trained TagGen using retrospective LR MR tagging images synthesized with R=3.3 undersampling.
• Evaluated TagGen against REGAIN (GAN-based super-resolution) on synthetic and prospective data.
• Prospectively acquired data using 10-fold acceleration (R=3.3 + GRAPPA-3).

Main Results:

• TagGen significantly outperformed REGAIN on synthetic data (p<0.05) for RMSE, PSNR, and SSIM.
• Radiologists rated TagGen superior to REGAIN for prospectively acquired 10-fold accelerated data (p<0.05).
• TagGen demonstrated improved tag grid quality, SNR, and overall image quality.

Conclusions:

• A diffusion-based generative super-resolution model (TagGen) was developed for MR tagging.
• TagGen integrates with parallel imaging for highly accelerated cine MR tagging.
• The method enhances tag grid quality in accelerated MR tagging acquisitions.