A neonatal apnoea monitor for resource-constrained environments

A prototype Android application was designed to monitor for apnoea in neonates using a smartphone. The application receives data from a wireless pulse oximeter and uses machine learning techniques to detect apnoea. Distribution of the system requires only the pulse oximeter and a current mid-range smartphone. This work builds on previous research, but with a particular focus on classifying events accurately using a reduced set of information appropriate to a resource-constrained environment. This information consists only of the photoplethysmogram (PPG) and a set of PPG-derived physiological variables including heart rate and respiration rate. Various methods using the Support Vector Machine (SVM) were assessed using data from 27 annotated stays in a neonatal intensive care unit, divided approximately in half into training and test data. The best approach was found to be a combination of a feature selection method based on mutual information and an SVM with a radial basis function kernel, producing a classifier with a sensitivity of 98.7%, a specificity of 62.2% and a balanced accuracy of 80.5% on a training set of 796 events, and a sensitivity of 76.9%, a specificity of 52.0% and a balanced accuracy of 64.4% on a test set of 663 events.