Modelling cerebrospinal fluid dynamics in a mouse brain

The flow of cerebrospinal fluid (CSF) in the in the brain plays an important role in clearance of waste and is linked to diseases such as Alzheimer’s and small vessel disease. We use physics informed neural networks (PINNs) to infer in vivo flow fields. This presentation describes various modelling and experimental efforts aimed at understanding the fluid dynamics of CSF flow. Specifically, we use PINNs to infer high-resolution pressure, volumetric flow rates, and shear stresses of CSF in tens-of-micron-scale channels from sparse two-dimensional (2D) velocity measurements and three-dimensional (3D) domain boundaries. We also use PINNs to infer brain-wide CSF flow from dynamic contrast-enhanced magnetic resonance imaging, which captures time-evolving concentration fields. Our approach includes the advection diffusion equation in the loss function, thus enforcing the physics. The neural network also learns the how the brain tissue permeability varies throughout the brain. We validate the approach using synthetic data from a realistic brain geometry with a realistic distribution of permeability values spanning several orders of magnitude, and we also show model predictions of flow and permeability from real data.