Accelerated cartilage T1ρ mapping with Denoising Diffusion Probabilistic Model (DDPM) and Generative Adversarial Network (GAN)

Motivation: MR quantitative T1ρ mapping has shown to be sensitive to early changes in diseases. However, the scan time is long due to the need for acquisition at several spin-lock times. Goal(s): This study aims to integrates a denoising diffusion model with a generative adversarial network to generate accelerated T1ρ-weighted images and quantitative maps. Approach: To enhance performance, the MRI physics model is incorporated into the sampling process of the diffusion model. Unlike supervised models, our method is independent of the subsampling patterns used in the acceleration process. Results: Our results demonstrate that our method can achieve superior multi-coil T1ρ-weighted images and T1ρ map. Impact: We studied the combination of DDPM with GAN for accelerated T1ρ imaging where both T1ρ-weighted images and T1ρ maps are obtained simultaneously from accelerated scans. The proposed method shows improvement over the competing methods.