Abstract

Motivation: Real diffusion-weighted MRI (DWI) has shown improved diffusion contrast and more accurate estimation of diffusion parameters. However, current real DWI is still not optimal and its performance highly depends on the choice of parameters for phase correction, reducing its robustness, especially in body DWI. Goal(s): To enable robust real DWI by improving phase correction and reducing noise floorApproach: We combined deep-learning based phase correction and deep learning reconstruction to enable robust phase correction and to reduce noise floor. Results: The phantom and healthy volunteer results demonstrated improved diffusion contrast in both acquired and synthesized high b-value neural and body DWI images. Impact: DL-based phase correction and image reconstruction enables robust real DWI imaging, improving diffusion contrast and quantitative measurements in both neuro and body. This technique can be used to study cancer staging and treatment response in both low and high field.