Machine-learning enabled wireless wearable sensors to study individuality of respiratory behaviors

• Published in: Biosensors & bioelectronics Vol. 173; p. 112799
• Main Authors: Chen, Ang, Zhang, Jianwei, Zhao, Liangkai, Rhoades, Rachel Diane, Kim, Dong-Yun, Wu, Ning, Liang, Jianming, Chae, Junseok
• Format: Journal Article
• Language: English
• Published: England Elsevier B.V 01.02.2021
• Subjects: Machine-learning; Posture recognition; Respiratory individuality; Respiratory monitoring; Wearable sensor; Respiratory individuality; Posture recognition; Respiratory monitoring; Wearable sensor; Machine-learning
• ISSN: 0956-5663; 1873-4235
• DOI: 10.1016/j.bios.2020.112799

Respiratory behaviors provide useful measures of lung health. The current methods have limited capabilities of continuous characterization of respiratory behaviors, often required to assess respiratory disorders and diseases. This work presents a system equipped with a machine learning algorithm, capable of continuously monitoring respiratory behaviors. The system, consisting of two wireless wearable sensors, accurately extracts and classifies the features of respiratory behaviors of subjects within various postures, wirelessly transmitting the temporal respiratory behaviors to a laptop. The sensors were attached on the midway of the xiphoid process and the costal margin, and 1 cm above the umbilicus, respectively. The wireless wearable sensor, consisting of ultrasound emitter, ultrasound receiver, data acquisition and wireless transmitter, has a small footprint and light weight. The sensors correlate the mechanical strain at wearing sites to lung volume by measuring the local circumference changes of the chest and abdominal walls simultaneously. Eleven subjects were recruited to evaluate the wireless wearable sensors. Three different random forest classifiers, including generic, individual, and weighted-adaptive classifiers, were used to process the wireless data of the subjects at four different postures. The results demonstrate the respiratory behaviors are individual- and posture-dependent. The generic classifier merely reaches the accuracy of classifying postures of 21.9 ± 1.7% while individual and weighted-adaptive classifiers mark substantially high, up to 98.9 ± 0.6% and 98.8 ± 0.6%, respectively. The accurate monitoring of respiratory behaviors can track the progression of respiratory disorders and diseases, including chronic respiratory obstructive disease (COPD), asthma, apnea, and others for timely and objective approaches for control. •Continuously monitor and extract features of respiratory behaviors completely and accurately.•Machine-learning enabled study of individuality of respiratory behaviors.•Postures recognition based on general respiration contributing to monitor posture-related respiratory diseases, like apnea.