Signal processing and classification for identification of clinically important parameters during neonatal resuscitation

Abstract

Neonatal mortality is a global challenge. One million newborns die each year within their first 24 hours as a result of complications during labour and birth asphyxia. Most of these deaths happen in low resource settings. However, basic resuscitation at birth can increase newborn survival. Identification of initial factors and simple therapeutic strategies determinant for neonatal outcome can aid health care workers provide the best follow-up during resuscitation. In this work, the initial condition of the newborn, the treatment given, and early heart rate response from manual bag mask ventilation are parameterized. The features are investigated in a machine learning framework to identify which features are determinant for the different outcomes. Using a selection of the defined features, an identification rate of 89% for newborns in the normal group, and an identification rate of 74% for episodes ending in death was found. This points to the direction of identifying the important factors of newborn survival.