CRNN with Bidirectional Frame-By-Frame Diffusion-Model-Based Refinement for Cardiac cine MRI Reconstruction

Motivation: In reconstruction of accelerated cardiac cine MRI, commonly used methods suffer from lack of details. Diffusion models have demonstrated outstanding ability of deblurring in existing studies, and potentially can recover details in MRI reconstruction. Goal(s): To develop a novel method leveraging diffusion model to refine the results of commonly used methods by recovering details. Approach: A bidirectional frame-by-frame refinement scheme of coarse CRNN result has been designed, and a 2D diffusion model with multi-scale feature extraction and fusion block incorporating temporal guidance has been proposed. Results: PSNR, SSIM and visualization of reconstruction results demonstrated the superiority of the proposed method over existing methods. Impact: A novel diffusion-model-based method of accelerated cardiac cine MRI reconstruction has been developed and improved the quality of reconstructed images. This may facilitate the application of accelerated cardiac cine MRI in clinical medicine, reducing patient discomfort and motion related artifacts.