Nonlinear Estimation of a Parsimonious Wiener Model for the Neuromuscular Blockade in Closed-Loop Anesthesia

The recursive identification of a parsimonious nonlinear Wiener model for the neuromuscular blockade in closed-loop anesthesia is considered. The performance of two popular nonlinear estimation techniques, namely the extended Kalman filter (EKF) and the particle filter (PF), is evaluated on synthetic and clinical data. The parameter estimates obtained with the PF, that does not rely on model linearization, exhibit less bias and shorter settling time than the ones produced by the EKF. This behavior persists when the parameter tracking capabilities of both estimation algorithms are assessed for the model in hand. Taking advantage of the model parameters that were recursively estimated from clinical data, it is demonstrated that the main source of intra-patient variability lies in the nonlinear pharmacodynamic part of the model. The distance to a bifurcation phenomenon leading to nonlinear oscillations of the Wiener model under PID feedback is also evaluated.