A machine learning method to explore the glymphatic system via poroelastodynamics

Abstract

Until recently, scientists thought that waste was cleared from the central nervous system primarily by diffusion, with waste slowly moving from the brain tissue toward the blood vessels. However, scientists have discovered a dedicated macroscopic waste clearance system, called glymphatic system, that performs efficient waste elimination through a unique system of perivascular channels formed by astrocytes, a type of glia. To better understand the complex dynamics of fluid movements in the glymphatic system, we implemented a four-compartmental poroelasticity model of the cerebral environment and used the model in a systematic and efficient parametric study aided by machine learning, a physiologically inspired perceptron method, to explore the functional impact of water transfer coefficients. The results suggested that the model captured the transport phenomenon of human brain. Moreover, within a specific distribution range of water transfer coefficients, the model indirectly predicts the existence of the glymphatic system, providing theoretical support to the glymphatic theory. Our study also demonstrates the feasibility of discovering physiological properties of complex biological systems through computer-aided modeling. Meanwhile, the presented novel method shown its potential for parametric study.