Identification of cardiac signals in ambulatory ECG data

Torabi, Mahdi

Abstract

The Electrocardiogram (ECG) is the primary tool for monitoring heart function. ECG signals contain vital information about the heart which informs diagnosis and treatment of cardiac conditions. The diagnosis of many cardiac arrhythmias require long term and continuous ECG data, often while the participant engages in activity. Wearable ambulatory ECG (AECG) systems, such as the common Holter system, allow heart monitoring for hours or days. The technological trajectory of AECG systems aims towards continuous monitoring during a wide range of activities with data processed locally in real time and transmitted to a monitoring centre for further analysis. Furthermore, hierarchical decision systems will allow wearable systems to produce alerts or even interventions. These functions could be integrated into smartphones.
A fundamental limitation of this technology is the ability to identify heart signal characteristics in ECG signals contaminated with high amplitude and non-stationary noise. Noise processing become more severe as activity levels increase, and this is also when many heart problems are present.
This thesis focuses on the identification of heart signals in AECG data recorded during participant activity. In particular, it explored ECG filters to identify major heart conditions in noisy AECG data. Gold standard methods use Extended Kalman filters with extrapolation based on sum of Gaussian models. New methods are developed using linear Kalman filtering and extrapolation based on a sum of Principal Component basis signals. Unlike the gold standard methods, extrapolation is heartcycle by heartcycle. Several variants are explored where basic signals span one or two heartcycles, and applied to single or multi-channel ECG data.
The proposed methods are extensively tested against standard databases or normal and abnormal ECG data and the performance is compared to gold standard methods. Two performance metrics are used: improvement in signal to noise ratio and the observability of clinically important features in the heart signal. In all tests the proposed method performs better, and often significantly better, than the gold standard methods. It is demonstrated that abnormal ECG signals can be identified in noisy AECG data.

Citation

Torabi, M. Identification of cardiac signals in ambulatory ECG data. (Thesis). University of Hull. https://hull-repository.worktribe.com/output/4223481

• Thesis Type: Thesis
• Deposit Date: Oct 6, 2021
• Publicly Available Date: Feb 24, 2023
• Keywords: Engineering
• Public URL: https://hull-repository.worktribe.com/output/4223481
• Additional Information: School of Engineering and Computer Science, The University of Hull
• Award Date: Dec 1, 2018