Continuous IoT-based maternal monitoring: system design, evaluation, opportunities, and challenges

Abstract

Maternal care encompasses health care services for pregnant women during pregnancy, childbirth, and the postpartum period. Maternity care providers aim to ensure a healthy pregnancy, safe delivery, and smooth transition to motherhood. Traditional maternal care is offered through regular check-ups by health care professionals.

In recent years, the emergence of Internet-of-Things (IoT)-based systems has transformed the way health care services are provided. These systems offer low-cost ubiquitous monitoring in everyday life settings and can be used for maternal monitoring. However, IoT-based maternal monitoring systems lack a comprehensive approach in maternal care because they are limited by sensing capabilities, specific health problems, and short periods of monitoring. Moreover, the use of IoT-based systems formaternal health monitoring requires addressing critical quality attributes, such as feasibility, energy efficiency, and reliability and validity of the collected physiological parameters. Quality assessment methods also must be integrated with such systems to discard the noisy part of collected parameters and improve the data quality. Furthermore, long-term, continuous IoT-based maternal monitoring by collecting data that was not traditionally available provides new opportunities, including analyzing the trend of physiological parameters during pregnancy and postpartum, as well as detecting maternal health issues.

This thesis presents an IoT-based maternal monitoring system and explores its potential in maternal care. We evaluate the system’s feasibility, reliability, and energy efficiency. We also discuss the practical challenges of implementing the system. Then, we validate the heart rate (HR) and heart rate variability (HRV) parameters that the system collects while the user is asleep and awake. In addition, we propose a deep-learning-based method for quality assessment of HR and HRV parameters to discard unreliable data and improve health decisions. We use the system to collect data from 62 pregnant women during pregnancy and three-months postpartum. Then, the reliable HR and HRV parameters are used to track the trends during pregnancy and postpartum.

Finally, we investigate maternal loneliness as a major mental health problem. We develop two predictive models to detect maternal loneliness during late pregnancy and the postpartum period. The models use the objective health parameters passively collected by the system and achieve high performance (weighted F1 scores > 0.87).