Understanding Brain Fluid Dynamics in Health and Disease Using Physics-Informed Machine Learning

Efficient cerebrospinal fluid circulation in the brain plays a vital role in clearing metabolic waste and maintaining neurological health, with impairments linked to neurodegenerative diseases. However, studying this circulation system remains challenging due to the lack of quantitative measurements of the flow throughout the brain. To address this, we model the brain as a porous medium governed by Darcy’s law, focusing on the estimation of velocity and permeability fields. We hypothesize that disruptions in the water channel proteins that decorate cell membranes alter the brain’s fluid flow. Using dynamic contrast-enhanced MRI (DCE-MRI) data, which records transient concentration fields of a contrast agent, we solve an inverse problem to recover the steady-state velocity and permeability fields. We develop a physics-informed neural network (PINN) that integrates the advection-diffusion equation, Darcy’s law, conservation of mass, and empirical measurements. The framework is first validated on synthetic data, then applied to real measurements from both healthy mice and mice genetically modified to lack aquaporin-4 water channels. Our results reveal distinct differences in inferred fluid dynamics between healthy and impaired brains, consistent with known biological mechanisms. This integrated modeling and inference approach not only enhances our understanding of the fluid transport in the brain but also offers a robust platform for future studies of brain clearance pathways in both health and disease.