Electronic Theses and Dissertations

Date

2021

Document Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Electrical & Computer Engineering

Committee Member

Mohd Hasan Ali

Committee Member

Frank Andrasik

Committee Member

Madhusudhanan Balasubramanian

Abstract

Artificial intelligence-enabled applications on edge devices have the potential to revolutionize disease detection and monitoring. Smartphones, with increased processing speed, storage capacity, and integrated sensors, hold tremendous promise in physiological sensing, monitoring, and development of Smart and Connected Communities (SCC). The major challenges are development of user-friendly low-cost sensing of physiological data, seamless data transfer while maintaining privacy, and data processing and inference at the edge while keeping the system performance at an expected level. The overall focus of my dissertation work is the development of a smart health (sHealth) framework that uses embedded artificial intelligence, such as edge computing, machine learning, etc., for disease early detection, severity estimation, and spatiotemporal monitoring for the SCC. Sleep Health evaluation is a prominent area where the advantages of edge-centric sHealth solutions can be leveraged. Rapid changes in socio-economic structure are impacting sleep health globally. Sleep apnea, insomnia, sleep deprivation, etc. are growing problems and impacting the health-related quality of life. To minimize the adverse health consequences, early detection and continuous monitoring of sleep disorders are beneficial. We investigated a minimalistic approach for the severity classification, severity estimation, and progression monitoring of obstructive sleep apnea (OSA) in the home environment using wearables. The recursive feature elimination technique was used to select the best feature set of 70 features from a total of 200 features extracted from polysomnograms. Then we used a multi-layer perceptron model to investigate the performance of OSA severity classification using a subset of features available from either Electroencephalography or Heart Rate Variability (HRV) and time duration of Oxygen saturation (SpO2 level). By using only computationally inexpensive features from HRV and SpO2, we were able to achieve an area under the curve of 0.91 and an accuracy of 83.97% can be achieved for the severity classification of OSA. For estimation of the apnea-hypopnea index, an accuracy of RMSE=4.6 and R-squared value=0.71 were achieved in the test set using only ranked HRV and SpO2 features. The Wilcoxon-signed-rank test indicated a significant change (p

Comments

Data is provided by the student.

Library Comment

Dissertation or thesis originally submitted to ProQuest

Share

COinS