Model-based Estimation of Ventricular Cerebrospinal Fluid Volume

This paper introduces a model-based estimator for ventricular cerebrospinal fluid (CSF) volume as a step toward improved hydrocephalus patient monitoring and therapy. For the development of this estimator, we propose a three-step pipeline: First, optimal experiment design is used to plan a safe medical intervention that maximizes the information observed about the patient's CSF dynamics. Second, the unknown parameters of a state-space model describing the CSF dynamics are identified through maximum likelihood estimation employing either expectation maximization or nonlinear programming. Third, a Kalman filter that uses intracranial pressure and bioimpedance measurements for online CSF volume estimation is constructed based on the identified CSF dynamics model. A custom-built measurement system and mechatronic test bench are used for in-vitro validation of the introduced methods. The model-based estimator achieves a 5.56% normalized root-mean-square estimation error during experimental testing. Smart shunt systems for hydrocephalus therapy could be augmented by the proposed estimator to allow for direct CSF volume regulation instead of pressure-driven CSF drainage control.