Electronic Theses and Dissertations

Date

2021

Document Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Electrical & Computer Engineering

Committee Chair

Dr. Chrysanthe Preza

Committee Member

Dr. Bashir Morshed

Committee Member

Dr. Hasan Ali

Committee Member

Dr. Aaron L Robinson

Committee Member

Dr. Amy L. De Jongh Curry

Abstract

In the future, Smart and Connected Communities (S&CC) will use distributed wireless sensors and embedded computing platforms to produce meaningful data that can help individuals, and communities. Here, we presented a scanner, a data reliability estimation algorithm and Electrocardiogram (ECG) beat classification algorithm which contributes to the S&CC framework .In part 1, we report the design, prototyping, and functional validation of a low-power, small, and portable signal acquisition device for these sensors. The scanner was fully tested, characterized, and validated in the lab, as well as through deployment to users homes. As a test case, we show results of the scanner measuring WRAP temperature sensors with relative error within the 0.01% range. The scanner measurement shows distinguish temperature of 1F difference and excellent linear dependence between actual and measured resistance (R2 = 0.998). This device hasdemonstrated the possibility of a small, low-power portable scanner for WRAP sensors.Additionally, we explored the statistical data reliability metric (DReM) to explain the quality of bio-signal quantitatively on a scale between 0.0 -1.0. As proof of concept, we analyzed the ECG signal. Our DReM prediction algorithm measures the reliability of the ECG signals effectively with low Root mean square error = 0.010 and Mean absolute error = 0.008 and coefficient of determination R2 value of 0.990. Finally, we tested our model against the opinions of three independent judges and presented R2 value to determine the agreement between judgments vs our prediction model.We concluded our contribution to the S&CC framework by analyzing ECG beat classification with a pipeline of classifiers that focuses on improving the models performance on identifying minority classes (ventricular ectopic beat, supraventricular ectopic beat). Moreover, we intended to minimize morphological distortion introduced due to indiscriminate use of filtering techniques on ECG signals. Our approach shows an average positive predictive value 95.21%, sensitivity of95.28%, and F-1 score 95.76% respectively.

Comments

Data is provided by the student.

Library Comment

Dissertation or thesis originally submitted to ProQuest

Notes

Open Access

Download

Share

COinS