Artificial intelligence velocimetry reveals in vivo pressure gradients in brain cerebrospinal fluid

Cerebrospinal fluid (CSF) circulates around and through brain tissue, sweeping away metabolic wastes that correlate with diseases including Alzheimer's, contributing to damaging swelling during stroke and cardiac arrest, and offering a potential pathway for drug delivery. Brain CSF flow is difficult to quantify in vivo at all, and quantifying the pressure gradients associated with flow has been altogether impossible up to now. We reveal pressure gradients for the first time by combining in vivo particle tracking with artificial intelligence velocimetry (AIV). In AIV, neural networks are trained to infer pressure and velocity fields by simultaneously minimizing mismatch with experimental measurements, minimizing error in the momentum and mass equations, and minimizing error in the boundary conditions. The pressure fields we infer extend throughout the three-dimensional perivascular spaces where CSF flows. We quantify typical pressure gradients and compare to prior estimates from theory. We show that the inertial term in the momentum equation is negligible, enabling lower-cost simulation in the future. With fluid-structure interaction in mind, we compare pressure fluctuations to the stiffness of bounding tissue. Finally, we characterize the magnitude and spatiotemporal variation of pressure gradients to gain insight into the fluid dynamical mechanisms driving CSF flow in the brain.