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

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