Wearable Soft Stethoscope With Noise Reduction Using Neural Network Towards Neonate Breath Monitoring

The long-term qualitative measurement of neonatal respiration has the potential to reduce the monitoring burden in neonatal wards and to predict changes in the condition of newborns. Since respiratory sounds provide not only respiratory rate but also qualitative data such as respiratory intensity, systems for temporal and qualitative measurement of respiration using wearable stethoscopes have been developed in recent years. However, their application to neonates has not been realized due to the fragile skin of neonates and the noisy environment in neonatal wards. In this study, a wearable stethoscope with a soft skin interface and a noise reduction algorithm using machine learning were combined to acquire breath sounds of newborns with cry noises removed. As a result, respiratory rate and heart rate calculated from the sound acquired by the wearable stethoscope tended to be less than the estimated values. On the other hand, the noise reduction algorithm using U-net Inspired convolutional neural network (CNN) improved the signal-tonoise ratio(SNR) of crying and breathing sounds by 26 dB.