Approaches for Improving Data Acquisition in Sensor-based MHealth Applications
Author | : Chunjong Park |
Publisher | : |
Total Pages | : 0 |
Release | : 2022 |
ISBN-10 | : OCLC:1367362770 |
ISBN-13 | : |
Rating | : 4/5 ( Downloads) |
Download or read book Approaches for Improving Data Acquisition in Sensor-based MHealth Applications written by Chunjong Park and published by . This book was released on 2022 with total page 0 pages. Available in PDF, EPUB and Kindle. Book excerpt: Mobile health (mHealth) applications enable people with little or no clinical experience to measure vital signs and screen for different health conditions with mobile devices. While such applications provide accessible health to the general population, recent mHealth applications leverage diverse sensors, require complex data acquisition procedures, and rely on complicated, black-box algorithms. Users are often uncertain about the quality of acquired data and resulting health-related predictions. They are also exposed to silent failures that lead to inaccurate results that could impact their medical decisions. To bridge the gap between underlying algorithms and non-expert users, this thesis explores approaches to design feedback for non-expert users to ensure that the screening algorithms only process high-quality data to provide accurate, reliable, and trustworthy results. Different feedback strategies should be applied at different stages of mHealth applications usage. During data acquisition, real-time, sensor-driven feedback is explored through two projects: (1) RDTScan, a smartphone-based rapid diagnostic test (RDT) reader, which employs real-time image quality assurance and guidance to capture high-quality RDT images, and (2) CapApp, a smartphone-based capillary refill time (CRT) assessment, which uses sensor-driven feedback to guide users to apply and release pressure for obtaining high-quality camera-based finger PPG signals. After data acquisition, the estimated uncertainty can inform users whether health algorithms can accurately process and interpret the acquired data. I explored leveraging state-of-the-art out-of-distribution detection methods for health ML models to improve the reliability of the results and user trust. By providing such information, users can only receive reliable predictions from the algorithms and only trust the diagnostic results with low uncertainty. This thesis provides a design space of feedback strategies to help non-expert users receive accurate and reliable health predictions from mHealth applications through these projects.