System identification for clinical diagnosis of hydrocephalus

摘要

Hydrocephalus is a neurological disorder which is associated with disturbed cerebrospinal fluid (CSF) system dynamics. Estimation of dynamical parameters is an important part of the diagnosis process, and can be performed via a controlled infusion of artificial CSF into the lumbar cavity. Current methods for testing and data analysis are not optimized in any way and may be very inaccurate. Maximizing information and minimizing experiment time are important for accuracy of the diagnosis, efficient use of hospital resources, and minimizing discomfort for the patient. In this paper, we show that a known and proven nonlinear differential equation model of the CSF dynamics can be transformed into a linear time invariant system via a nonlinear change of variables. After this change of variables, the parameter estimation problem becomes a standard system identification problem. We address important issues such as model validation, prefiltering and disturbance modelling. We present experimental results on a phantom, as well as preliminary data from a clinical trial currently in progress.