Accelerated Prediction of Bradycardia in Preterm Infants Using Time-Frequency Analysis

The rapid growth of the high-performance data processing algorithms and fusion of wearable technologies has enabled us to continuously monitor the health status of infants. However, monitoring preterm infants is still a challenge due to their frightfully tiny size and their undeveloped skin. In this paper, we have demonstrated a framework of complete monitoring of the preterm infants and real-time accelerated prediction of bradycardia. Real-time prediction of bradycardia episodes in the NICU has the potential to provide quality care to these neonates. In the proposed system we incorporated a multi-GPU gradient boosting algorithm that was able to outperform the traditional CPU’s performance. This can overcome the manually maintained progress reports by nurses, which is a major hurdle in the NICU. The system maps the workflow in a Java based responsive application to provide statistical information along with growth charts and reports. The system extracts data with a specific interval to feed the GPU based extreme gradient boosting model. The feature extraction was performed on the time and frequency domain of the heart rate of infants to predict an episode of bradycardia. With an average accuracy of 86% and shortest detection time when compared to the models of other similar products, the proposed system showed that it can improve care time, minimize the skill gap, analyze early disease perdition, and reduce preterm infant’s morbidity and mortality.