Motivation: Realistic MR signal simulations incorporating microvascular structures and water-diffusion are crucial to understand fMRI BOLD or DSC experiments and are directly applicable in MRF techniques. Yet, they remain computationally challenging. Goal(s): To propose a novel simulation tool that efficiently accounts for magnetic susceptibility and water-diffusion effects and can simulate MR-signals from 3D realistic microvasculatures in seconds. Approach: MR-WAVES combines fast Bloch-simulations, local frequency distributions, and deep neural network water-diffusion predictions. Results were validated against standard microvascular simulations and applied to MRvF estimates in healthy rat brains. Results: MR signal generation was accelerated nearly 13,000-fold compared to standard simulations, while maintaining MRF reconstruction accuracy. Impact: We propose a deep learning-based simulation tool for rapid MR signal generation that accounts for microvascular susceptibility and water diffusion effects. This tool could enhance our understanding of the BOLD effect and improve microvascular parameter quantification via MRF methods.
Coudert et al. (Tue,) studied this question.