Detection of Bradycardia in Preterm Infants by Using ECG and Respiratory Signals

Abstract

This study is conducted to detect bradycardia in preterm infants. The signal of electrocardiogram (ECG) has been widely applied to the detection. We propose to consider respiratory signals to improve the accuracy of classification. The machine learning model AutoML is used for feature selection and classification. The training data include ECG and respiratory signals. The target is to determine whether symptoms of bradycardia occur in preterm infants. Through the experimental results, the classification by analyzing the features of the ECG together with the respiratory signal showed an average accuracy of 79.2% which was better than 75.32% by using ECG only and 62.08% by using a respiratory signal only. The comparison result shows that in addition to ECG, respiratory signals are important in the detection of bradycardia.