Detection of Asphyxia and Apnea of Newborns Under Incubator Treatment Using Respiration Signals and Machine Learning

Abstract

Premature infants may have several health problems, breathing problems are the most common form of these problems. Detection of breathing problems in newborns accurately is a very important issue. Premature infants generally are taken in neonatal intensive care under incubator treatment. This study aimed to propose a machine learning-based algorithm using respiration signals of infants to detect abnormal respiration events. 9 Time domain features were extracted from respiration signals and used as inputs of classifiers. The neighborhood component analysis method was applied to detect the most important features. 4 Features were selected as important features and were given as inputs of classifiers. The K-Nearest Neighbors KNN (K=9) algorithm reached the best performance when 2 different feature sets were given as inputs. The accuracy of KNN (K=9) with 9 features was 92.05%+/- 8.02% while with 4 features was 90.68%+/- 2.11%.