Development of a machine learning model to identify intraventricular hemorrhage using time-series analysis in preterm infants

Abstract

Abstract Although the prevalence of intraventricular hemorrhage (IVH) has remained high, no optimal strategy has been established to prevent it. This study included preterm newborns born at a gestational age of < 32 weeks admitted to the neonatal intensive care unit of a tertiary hospital between January 2013 and June 2022. Infants who had been observed for less than 24 h were excluded. A total of 14 features from time-series data after birth to IVH diagnosis were chosen for model development using an automated machine-learning method. The average F1 scores and area under the receiver operating characteristic curve (AUROC) were used as indicators for comparing the models. We analyzed 778 preterm newborns (79 with IVH, 10.2%; 699 with no IVH, 89.8%) with a median gestational age of 29.4 weeks and birth weight of 1180 g. Model development was performed using data from 748 infants after applying the exclusion criteria. The Extra Trees Classifier model showed the best performance with an average F1 score of 0.93 and an AUROC of 0.999. We developed a model for identifying IVH with excellent accuracy. Further research is needed to recognize high-risk infants in real time.

Keywords

• Machine learning
• Cerebral intraventricular hemorrhage
• Neonatal intensive care