Publication
Over the last two decades, Diffusion-Weighted Magnetic Resonance Imaging (DW-MRI) has been developed as a powerful MR imaging modality that allows estimating non-invasively the microscopic structure of biological tissues by exploiting the natural motion of water molecules undergoing thermal agitation, i.e., Brownian motion. Today, this advanced MRI modality has been widely used in research to study brain organization, i.e., brain connectivity and the brain tissue microstructure.
This project aims to tackle current microstructure imaging limitations and extend this work to validate state-of-the-art methods. The current approaches rely on analytical models of the microstructure-MR signal relationship. They use simplifying assumptions about the tissue microstructure geometry and physical properties to keep this relationship tractable. In this project, we will tackle those limitations by developing an entirely new simulation-based framework. We will start with the microscopic geometrical and physical modeling of the white matter to create detailed and realistic synthetic substrates. Then, we will use and extend our numerical simulator to obtain DW-MRI signals corresponding to the generated synthetic substrates for selected DW-MRI acquisition protocols. Next, since we know the ground truth microstructure properties of the numerical substrates, we will study microstructure parameters estimation from the simulated DW-MRI signal using models and simulated-assisted machine learning. Afterward, it will be possible to assess the specificity and the sensitivity of the existing and newly developed methods for a specific MR sequence. Our methods will be optimized through iterations of this multi-step approach: 1) synthetic tissue generation 2) parameter estimation and sensitivity analysis, 3) MR sequence parameter optimization and simulation, and 4) validation and applications on real-tissues.
This project aims to enhance the Monte-Carlo simulations existing in the literature. We aim to evaluate the specificity and sensitivity of feature estimation from White Matter models used to study tissue composition and integrity. To do so, we are performing analyses of substrates that will have axon-line features. Moreover, by the end of this thesis, we will provide novel methods and tools helpful in assessing and validating microstructure imaging technics.
Keywords: medical imaging, microstructure imaging, diffusion MRI, inverse problems, brain imaging, white matter, Monte Carlo simulations, realistic substrates.