Predicting Onset of Bradycardia in Premature Infants Using ECG Signals

This talk will present an onset bradycardia detection and prediction technique that only relies on ECG signals. It consists of a novel algorithm based on the notion that the cardiovascular system can be treated as a dynamic system, and that under bradycardia, this dynamic system reacts abnormally due to the temporal and spatial destabilization. Beat signatures were derived from the ECG signal using peak detection techniques, and these signatures were used to estimate the pole of the cardiovascular system in the complex domain. A pole is defined as a complex number that represents the natural frequency and shape of a beat. The location of the poles on the complex plane reveals the state of the cardiovascular system and shows the trend when a system migrates from normal to bradycardia state. Based on this information, a warning can be generated. We will discuss this novel bradycardia detection and prediction approach in detail and present the result from our proof-of-concept study using data collected from ten preterm infants.
A major advantage of our algorithm is that it uses only a single channel ECG signal and requires much less training time, while performs as well as or better than the state-of-the-art prediction methods. Unlike conventional methods, our algorithm relies on changes in inherent dynamics of the cardiovascular system to predict the occurrence of bradycardia. For example, the point-process method, which is one of the latest methodologies in early detection of onset of bradycardia, shows very similar results compared to our algorithm, with a mean Area Under the Curve (AUC) of the receiver operating characteristic (ROC) curve of 0.79 for the same ten infant datasets. The AUC values ranged between 0.72 and 0.93 for the point-process method, and between 0.70 to 0.92 for our algorithm. However, the point-process method requires extensive training data of between 6-18 hours, while our algorithm only requires the first few minutes of ECG data to calibrate our system model. In terms of prediction, our algorithm can predict the on-set bradycardia with more than 45 s of lead time, providing crucial warning for bradycardia treatment of preterm infants.