Abstract

Cardiac signals provide variety of information related to the patient's health. One of the most important is for medical experts to diagnose the functionality of a patient’s heart. This information helps the medical experts monitor heart disease such as atrial fibrillation and heart failure. Atrial fibrillation (AF) is one of the most major diseases that are threatening patients’ health. Medical experts measure cardiac signals usng the Electrocardiogram (ECG or EKG), the Photoplethysmogram (PPG), and more recently the Videoplethysmogram (VPG). Then they can use these measurements to analyze the heart functionality to detect heart diseases. In this study, these three major cardiac signals were used with different classification methodologies such as Basic Thresholding Classifiers (BTC), Machine Learning (SVM) classifiers, and deep learning classifiers based on Convolutional Neural Networks (CNN) to detect AF. To support the work, cardiac signals were acquired from forty-six AF subjects scheduled for cardioversion who were enrolled in a clinical study that was approved by the Internal Review Committees to protect human subjects at the University of Rochester Medical Center (URMC, Rochester, NY), and the Rochester Institute of Technology (RIT, Rochester, NY). The study included synchronized measurements of 5 minutes and 30 seconds of ECG, PPG, VPG 180Hz (High-quality camera), VPG 30 Hz (low quality webcam), taken before and after cardioversion of AF subjects receiving treatment at the AF Clinic of URMC. These data are subjected to BTC, SVM, and CNN classifiers to detect AF and compare the result for each classifier depending on the signal type. We propose a deep learning approach that is applicable to different kinds of cardiac signals to detect AF in a similar manner. By building this technique for different sensors we aim to provide a framework to implement a technique that can be used for most devices, such as, phones, tablets, PCs, ECG devices, and wearable PPG sensors. This conversion of the different sensing platforms provides a single AF detection classifier that can support a complete monitoring cycle that is referring to screen the patient whether at a hospital or home. By using that, the risk factor of heart attack, stroke, or other kind of heart complications can be reduced to a low level to prevent major dangers, since increasing monitoring AF patients helps to predict the disease at an early stage as well as track its progress. We show that the proposed approach provides around 99% accuracy for each type of classifier on the test dataset, thereby helping generalize AF detection by simplifying implementation using a sensor-agnostic deep learning model.

Library of Congress Subject Headings

Atrial fibrillation--Diagnosis; Deep learning (Machine learning); Multisensor data fusion; Hospital care

Publication Date

12-2021

Document Type

Thesis

Student Type

Graduate

Degree Name

Electrical Engineering (MS)

Department, Program, or Center

Electrical Engineering (KGCOE)

Advisor

Gil Tsouri

Advisor/Committee Member

Ferat Sahin

Advisor/Committee Member

Jamison Heard

Campus

RIT – Main Campus

Plan Codes

EEEE-MS

Share

COinS