Multi-dimensional denoising of diffusion MRI using low rank tensor dictionary learning

Motivation: Joint denoising diffusion-weighted images in high b-valuesGoal(s): to achieve a better joint denoising of diffusion-weighted image at high b-values. Approach: A low-rank tensor dictionary learning model was introduced to joint denoising diffusion-weighted images at b-values of 1000/2000 s/mm2 for both simulations and in vivo datasets. Three state-of-the-art methods, MPPCA, WNNM, and a pre-trained complex-valued DnCNN were compared. Results: In the simulation, the proposed method achieved the smallest RMSE. In vivo, the proposed method demonstrated a promising denoising ability while better preserving the image structures. Impact: We proposed a low-rank tensor dictionary learning method to better exploit non-local spatial redundancies and image correlations across different b-values with learned dictionaries and low-rank tensor approximation, providing a promising performance in denoising and preserving the image structures.