Identification of multiple-input, linear and nonlinear, time-varying systems and binary response systems for biomedical applications

System identification is the process of building dynamical models from measured data in order to determine and quantify the underlying relationships between them. Ideally, the resulting mathematical models ought to imitate precisely the observed behavior of the system under examination. In this doctoral dissertation, we focus on developing effective methodologies for quantifying dynamic interrelationships in physiological systems using parametric, nonparametric and connectionist approaches. Due to the complex nature of physiological functions, standard system identification methods, which usually assume linear and time-invariant interrelationships, fail. Thus, this work describes fast and reliable modeling schemes that are capable of dealing with (a) multiple input systems (b) nonlinear dynamics (c) nonstationarities in system dynamics and (d) binary responses. These schemes were applied in combination with Laguerre-Volterra (LV) models, which can capture a wide range of nonlinear dynamic input-output causal interrelationships and Multivariate Autoregressive models (MVAR), which are used to detect couplings and causality between time series. The performance of the abovementioned methodologies was assessed using both simulations and experimental data. Specifically, we examined,•The time-varying (TV) characteristics of Cerebral Autoregulation (CA) in patients suffering from Vasovagal Syncope (VVS) during Head-Up Tilt (HUT) testing. •Exercise-induced cardiovascular and cerebrovascular changes in healthy subjects and stroke survivors.•Neuronal responses to subthalamic nucleus (STN) Local Field Potentials (LFP) in Parkinson's Disease (PD) patients undergoing Deep Brain Stimulation (DBS).