In this work, three different black box models were designed to model the blood oxygen saturation (SpO2) of a neonatal infant. The inputs into each of the models where fraction of inspired oxygen, heart rate, and respiratory rate. The three modeling types used were fuzzy logic, neural network, and a transfer function model. Each model was trained on a window of data. The models were tuned using a design of experiments and the optimum window size was found for each model. The models were tested to see how long the model adequately represents the future SpO2. The best model was found to be the transfer function model which adequately modeled the future SpO2 for an average of 51.8 seconds.

This content is only available via PDF.
You do not currently have access to this content.