Real-Time Multi-Level Neonatal Heart and Lung Sound Quality Assessment for Telehealth Applications

Abstract

In this study, a new method is proposed to assess heart and lung signal quality objectively and automatically on a 5-level scale in real-time, and to assess the effect of signal quality on vital sign estimation. A total of 207 10 s long chest sounds were taken from 119 preterm and full-term babies. Thirty of the recordings from ten subjects were obtained with synchronous vital signs from the Neonatal Intensive Care Unit (NICU). As a reference, seven annotators independently assessed the signal quality. For automatic quality classification, 400 features were extracted from the chest sounds. After feature ranking and selection, class balancing, and hyperparameter optimization, a variety of multi-class and ordinal classification and regression algorithms were trained. Then, heart rate and breathing rate were automatically estimated from the chest sounds. For the deep learning model, YAMNet, a deep convolutional neural network pre-trained on the AudioSet-Youtube corpus for sound classification was used. After modification of the final output layers of the neural network and class balancing, transfer learning was applied to YAMNet for heart and lung signal quality classification. The results of subject-wise leave-one-out cross-validation show that the best-performing models had a balanced accuracy of 56.8% and 51.2% for heart and lung qualities, respectively. The best-performing models for real-time analysis (< 200 ms) had a balanced accuracy of 56.7% and 46.3%, respectively. Our experimental results underscore that increasing the signal quality leads to a reduction in vital sign error.

Keywords

• Breath sound
• deep learning
• heart rate
• heart sound
• neonatal monitoring
• ordinal regression