Analysis, modelling and control of oxygen therapy for premature infants

Premature babies are the babies having a gestinal age of less than 37 weeks. These babies due to their early birth suffer from respiratory problems mainly because of the immaturity of some organs, namely the brain and the lungs. Chemoreceptors present in the brain and the trachea act as sensors and determine how much air a human needs to breathe. These sensors use carbon dioxide and oxygen concentrations present in the body to make their decisions. However, since the brain of premature babies may not be fully developed, the efficiency of these chemoreceptors can be significantly low. Moreover, the lungs themselves are sometimes not properly developed to allow for efficient gas exchange. As a consequence, premature babies often have irregular breathing patterns. The most serious case is the occurrence of apneas during which breathing stops completely for a certain period of time. In the event of an apnea, saturation of oxygen in the blood drops significantly. Different techniques are used to counter this drop and one of the medical interventions is oxygen therapy during which supplemental oxygen is externally supplied through the nasal cavity. The concentration of oxygen (FIO2) supplied to the baby can be varied where the lowest percentage is 21% (air). In neonatal intensive care units (NICU), dedicated nurses monitor the oxygen saturation level in the arterial blood of the babies which is non-invasively measured with an oximeter. If the saturation level is too low, nurses would normally increase the FIO2 level in an attempt to achieve the desired saturation level. However, this approach can be risky since there is a safe limit for the saturation level and the amount of oxygen that can be supplied. A very high oxygen concentration can be harmful, leading to blindness and other long term complications. This is known as oxygen toxicity. This thesis proposes a solution to deal with this issue that was firstly tested through simulation. A mathematical model for the respiratory system of a preterm infant was previously developed in Matlab by a former PHD student from Monash University, Scott Sands. This model was used as a starting point for this research and was further modified to investigate the proposed solution. Two control strategies, with arterial oxygen saturation as input are used together to determine the FIO2 required during oxygen therapy; one is a PI control and the other is a feedforward control. The PI control takes care of small fluctuations in oxygen saturation level and maintains it at a reference level of 94%. An apnea cannot be predicted and hence they are only known after it has occurred. The feedforward control is used to build a reserve of oxygen in the lungs so that when a chain of apneas occurs, there is enough oxygen in the lungs to support the metabolic activities of the body. The feedforward control does so by increasing FIO2 to a constant value for a pre calculated period of time when an apnea occurrence has been detected. Even though the oxygen saturation in the blood will drop, this control reduces the fall to a level which is considered safe. The PI controller is switched off during apneas since the difference between the reference and the input is too large for it to function. The unavoidable presence of noise in real life has also been taken into account and together with white Gaussian noise, a filter has been added in the model. Also, a prediction mechanism has been introduced to reduce the possibility of noisy input signals being detected as apneas. The combination of the above techniques has resulted in satisfactory test results in simulation studies. Building upon the simulation results, a study has been carried out to test the proposed feedforward controller. In the study, a lamb was forced into periods of breathing and apneas and FIO2 and applied PEEP were used as control variables. PEEP is an applied pressure against exhalation. Both variables were kept constant throughout an initial experiment. In the following experiments, one of the variables was increased and its effect on oxygen saturation was observed. Firstly, the effect of increasing FIO2 was investigated at a constant PEEP value. In another set of experiments, the effect of increasing PEEP was investigated. The same experiments were then repeated for apneas having a longer duration. It was found that increasing FIO2 or applied PEEP during apneic episodes results in a decrease in arterial oxygen desaturation. If both PEEP and supplemental oxygen are used, the FIO2 required is reduced and therefore the risk of oxygen toxicity is decreased. Also, as expected, increasing the duration of the apneas resulted in the need for more supplemental oxygen and applied PEEP to prevent oxygen saturation from falling to dangerously low values. Finally a microcomputer based electronic controller has been built to implement the proposed solution. The electronic controller has been tested offline using the recorded data of oxygen therapies conducted on premature babies at Monash Medical Centre. In the absence of real-time signal, the controller detects apneic signals as expected and reacts by increasing FIO2 in a manner similar to the simulation. The FIO2 calculated by the controller controls the servo motor mounted on a ventilator, which in turns moves the ventilator control knob to a position reflecting the FIO2 determined by the controller. Testing the controller in real time is to be another project once ethical approval is obtained.