Parameterized uncertainty modeling applied to the oxygen control of neonates with lung disease

Abstract

[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] Premature infants born in a hospital's Neonatal Intensive Care Unit (NICU) commonly have underdeveloped lungs which is a cause of Respiratory Distress Syndrome (RDS). In this instance the child must be placed on an assisted breathing apparatus. A mechanical ventilation or increased inspired oxygen assists the neonate's ability to maintain healthy blood oxygen levels. For prematurely born infants, a desired range of 85%-92% blood oxygen saturation is healthiest. Current methods of regulating blood saturation levels are through manual regulation by human interaction. A nurse in the NICU must be on hand to closely monitor vital signs of the infant and manually adjust the percentage of inspired oxygen accordingly. An automatic oxygen control system could yield higher accuracy and increased reliability and security while simultaneously reducing the workload of nurses in the NICU. Clinical data of saturation rates of neonates has been collected and parameterized into first order, frequency-based transfer functions. The parameters of the systems are: DC Gain, Time Constant, and Time Delay. Each oxygen saturation event has its own values to the parameters. Using a genetic algorithm, a lower dimension uncertainty model can be optimized to the spectrum of parameters, representing the variable range of oxygen saturation rates. The model of the transfer functions' coefficients is used to resample a set of data dependent on the original data's density. The lower dimensional parameterization and the increased sample size lowers the model uncertainty, and for which a less conservative controller can designed to robustly and accurately regulate the blood oxygen level of premature infants.