Cardiovascular parameter estimation using a computational model

Abstract

Modern intensive care units are equipped with a wide range of patient monitoring devices, each continuously recording signals produced by the human body. Currently, these signals need to be interpreted by a clinician in order to assess the state of the patient, to formulate physiological hypotheses, and to determine treatment options. With recent technological advances, the volume of relevant patient data acquired in a clinical setting has increased. This increase in sheer volume of data available, and its lack of organization, have rendered the clinical decision-making process inefficient. In some areas, such as hemodynamic monitoring, there is enough quantitative information available to formulate computational models capable of simulating normal and abnormal human physiology. Computational models tend to synthesize information in one common framework, thereby improving data integration and organization. Through tuning, such models could be used to track patient state automatically and to relate properties of the observable data streams directly to the properties of the underlying cardiovascular system. In our research efforts, we implemented a pulsatile cardiovascular model and attempted to match its output to simulated observable hemodynamic signals in order to estimate cardiovascular parameters.

(cont.) Tracking model parameters in time reveals disease progression, and hence it can be very useful for patient-monitoring purposes. As the observable signals are generally not rich enough to allow for the estimation of all the model parameters, we focused on estimating only a subset of parameters. Our simulations indicate that observable data at intra-beat timescales can be used to estimate distending blood volume, peripheral resistance, and end-diastolic right compliance to reasonable degrees of accuracy. Furthermore, our simulation results based on a real patient hemorrhage case reveal that clinically significant parameters related to bleeding rate and peripheral resistance can be tracked reasonably well using observable patient data at inter-beat timescales.