Microstructure.jl: a Julia Package for Probabilistic Microstructure Model Fitting with Diffusion MRI

Motivation: Compartment models can provide biologically meaningful microstructure maps. However, lack of robustness, evaluation and speed in parameter estimation hinder their interpretability and wider applications. Goal(s): 1) A flexible, extendable and coherent framework for modelling microstructure; 2) probabilistic estimators with improved fitting evaluation and computation speed. Approach: We use the high-performance and intuitive programming language Julia and machine learning fitting methods to develop the tool. We demonstrate neural network estimators for various models on publicly available datasets. Results: Parameter estimation and uncertainty quantification takes less than a minute per whole-brain dataset on single CPU. Further evaluations of fitting on synthetic datasets enhances interpretation. Impact: An easy-to-use tool is provided to the community for microstructural mapping from diffusion MRI across biophysical models with high computation speed, fitting evaluation and uncertainty quantification, applicable to data feasible on typical research and higher-performance scanners.