Connecting Diffusion MRI to Skeletal Muscle Microstructure: Leveraging Meta-Models and GPU-acceleration

Due to its non-invasive nature, diffusion-weighted MRI (dMRI) has shown promise as a method to quantify skeletal muscle's microstructure; however, connecting the dMRI signal of muscle to the underlying microstructure is difficult. Numerical models of dMRI can parameterize this relationship, but the associated computational expense has prohibited extensive use. Here, efficient numerical methods are presented to address this problem. In particular, a meta-model representation of a lattice Boltzmann model of dMRI is formulated and shown to be both accurate and several orders of magnitude faster to evaluate. It is also demonstrated how such a meta-model can help inform dMRI pulse profile selection in encoding microstructural information into the dMRI signal. Additionally, histologically-informed simulations are performed, allowing comparison of the numerical model's simplified parameterization with the more complex topology of skeletal muscle. Finally, an efficient inversion method is proposed to infer microstructural parameters of muscle from dMRI signal using a GPU-accelerated numerical model. The inversion method is able to infer microstructural parameters from dMRI signal when the underlying geometry matches the numerical model's, however, the simplified numerical model does not agree with simulations of more complex muscle tissue.