Jian NiFollow

Award Date

May 2023

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical and Computer Engineering

First Committee Member

Yingtao Jiang

Second Committee Member

Mei Yang

Third Committee Member

Ebrahim Saberinia

Fourth Committee Member

Hui Zhao

Number of Pages

109

Abstract

Due to the rapid development of computing and sensing technologies, Internet of Things (IoT)-based cardiac monitoring plays a crucial role in providing patients with cost-efficient solutions for long-term, continuous, and pervasive electrocardiogram (ECG) monitoring outside a hospital setting. In a typical IoT-based ECG monitoring system, ECG signals are picked up by sensors located on the edge, and then uploaded to the remote cloud servers. ECG interpretation is performed for the collected ECGs in the cloud servers and the analysis results can be made instantly available to the patients as well as their healthcare providers.In this dissertation, we first examine the ECG classification models in the cloud. Although deep learning technologies have shown their great use in extracting ECG signal features and recognizing useful patterns for diagnosis, the existing methods are found to have unacceptable levels of performance (with an accuracy capped at only 60%) in identifying certain abnormal rhythms that can cause life-threating cardiac events. In combating this deficiency, we have developed three methods that help produce, preserve, and sharpen the abnormality-relevant features needed to improve the detection of the abnormalities. These three methods are then integrated into a DNN framework for the detection of the ECG rhythms of interest. The experiment results on a publicly available data set demonstrate the effectiveness of the proposed method with the best accuracy result ever published. On the edge end of the IoT-based ECG monitoring system, both extremely noisy and almost noise-free ECGs could be locked in by the device worn by a mobile patient. However, transmitting an indiscriminate collection of noisy and noise-free ECG cycles to the cloud for the categorization of cardiac abnormalities typically leads to significant false alarm rates. Alternatively, merely relying on a single denoising or quality assessing process on the edge to cope with all the recorded ECG signals can also be problematic, as the former can catastrophically distort those noise-free sections of the ECG signal, while the latter tends to cause notable loss of meaningful clinical information by discarding the signal sections that stand a good chance to be recovered by a denoising process. In this dissertation, we present a series of machine learning based models in support of edge-level stratification and preprocessing, for selecting the ECG signals that either have clear morphologies or retain their morphologies after necessary denoising to upload to the cloud. On the other hand, signals that are useless for diagnosis will be deleted early in the signal chain to lessen the load that would otherwise be imposed on the communication network and the cloud. In specific, the severity of the noise presence in the collected ECG signals is first evaluated right on the edge, after which the ECG signals get stratified into three levels and processed accordingly: (1) Signals that are assessed to have clear morphologies are admitted to the cloud for classification; (2) Signals with significantly corrupted morphologies—caused by baseline wandering, electrode motion, and muscle artifacts—are judged to be useless for classification on the cloud and are therefore dropped right away on the edge; (3) Signals that fall between the previous two extremes with partially corrupted morphologies are warranted to go through a denoising process. This very last type of signals after denoising will be assessed again by a dedicated quality assurance algorithm, and only the denoised signal that carries recognizable diagnostic information will be sent to the cloud for classification. The performances of the proposed method are evaluated using five publicly available datasets, and the results have confirmed a saving of the network traffic and a noticeable load reduction at the cloud, which is critical to an edge-cloud computing environment. Since selective denoising, indicated above, becomes an integral part of ECG processing flow of the proposed (IoT)-based cardiac monitoring system, we have developed a set of machine-learning based denoising models that take into account of the limited power capabilities of the edge. Instead of resting on sophisticated and power-hungry denoising methods to indiscriminately cleanse ECG signals across the whole spectrum of noise conditions, our focus is placed on denoising ECG signals that have moderate noise levels and thus being able to recover useful ECG morphologies for ECG signal stratification purposes described above. Specially, we propose a series of machine-learning based denoising models that allows us (1) to select signals’ spectrums that are most relevant to diagnostically useful morphologies in the frequency domain, and subsequently, (2) recover recognizable diagnostic information from them. The experiment results on five publicly available datasets confirm the effectiveness of the proposed method.

Keywords

Deep learning; Electrocardiogram monitoring; Internet of Things

Disciplines

Artificial Intelligence and Robotics | Computer Engineering | Engineering | Health and Medical Administration

File Format

pdf

Degree Grantor

University of Nevada, Las Vegas

Language

English

Rights

IN COPYRIGHT. For more information about this rights statement, please visit http://rightsstatements.org/vocab/InC/1.0/

Download


Share

COinS