A Wide-bandwidth Nanocomposite-Sensor Integrated Smart Mask for Tracking Multi-phase Respiratory Activities for COVID-19 Endemic

Abstract

A global sentiment in early 2022 is that the COVID-19 virus could become endemic just like common cold flu viruses soon. The most optimistic view is that, with minimal precautions, such as vaccination, boosters and optional masking, life for most people will proceed as normal soon. However, as warned by A. Katzourakis of Oxford University recently, we must set aside lazy optimism, and must be realistic about the likely levels of death, disability and sickness that will be brought on by a 'COVID-19' endemic. Moreover, the world must also consider that continual circulation of the virus could give rise to new variants such as the new BA.2 variant (a subvariant of Omicron) continues to spread across the US and parts of Europe. Data from the CDC is already showing that BA.2 has been tripling in prevalence every two weeks. Hence, globally, we must use available and proven weapons to continue to fight the COVID-19 viruses, i.e., effective vaccines, antiviral medications, diagnostic tests and stop an airborne virus transmission through social distancing, and mask wearing. For this work, we have demonstrated a smart mask with an optimally-coupled ultra-thin flexible soundwave sensors for tracking, classifying, and recognizing different respiratory activities, including breathing, speaking, and two-/tri-phase coughing; the mask's functionality can also be augmented in the future to monitor other human physiological signals. Although researchers have integrated sensors into masks to detect respiratory activities in the past, they only based on measuring temperature and air flow during coughing, i.e., counting only the number of coughs. However, coughing is a process consisting of several phases, including an explosion of the air with glottal opening producing some noise-like waveform, a decrease of airflow to decrease sound amplitude, and a voiced stage which is the interruption of the air flow due to the closure of glottal and periodical vibration of partly glottis, which is not always present. Therefore, sensors used for cough detection should not be only sensitive to subtle air pressure but also the high-frequency vibrations, i.e., a pressure sensor that needs to be responsive to a wide input amplitude and bandwidth range, in order to detect air flows between hundreds of hertz from breath, and acoustic signals from voice that could reach ~ 8000 Hz. Respiratory activities data from thirty-one (31) human subjects were collected. Machine learning methods such as Support Vector Machines and Convolutional Neural Networks were used to classify the collected sensor data from the smart mask, which show an overall macro-recall of about 93.88% for the three respiratory sounds among all 31 subjects. For individual subjects, the 31 human subjects have the average macro-recall of 95.23% (ranging from 90% to 100%) for these 3 respiratory activities. Our work bridges the technological gap between ultra-lightweight but high-frequency response sensor material fabrication, signal transduction and conditioning, and applying machining learning algorithms to demonstrate a reliable wearable device for potential applications in continual healthy monitoring of subjects with cough symptoms during the eventual COVID-19 endemic. The monitoring and analysis of cough sound should be highly beneficial for human health management. These health monitoring data could then be shared with doctors via cloud storage and transmission technique to help disease diagnosis more effectively. Also, communication barriers caused by wearing masks can be alleviated by combining with the speech recognition techniques. In general, this research helps to advance the wearable device technology for tracking respiratory activities, similar to an Apple Watch or a Fitbit smartwatch in tracking physical and physiological activities.