In Diffusion MRI, the large variability of acquisition schemes limits broader use of Deep Learning for parameter estimation, with data-specific models needed for high-quality predictions. To reduce dependency on training data, we present dtiRIM, a recurrent neural network that learns a regularized solution to a model-based inverse problem. Using the diffusion model allows independent parameters (e.g. gradient directions) to also influence the estimation. We show that a single dtiRIM model predicts diffusion parameters for multiple datasets with lower error than the current state-of-the-art. Our study suggests that dtiRIM has the potential to be the first general learning method for DTI.
This abstract and the presentation materials are available to members only; a login is required.