Jain, Utkars
(2024)
Utilizing Machine Learning to Automatically Determine Waveform Markers Within Electrocardiograms.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
Electrocardiograms (ECGs) are an inexpensive, non-invasive, and widely available cardiovascular test that measures the electrical beats and rhythms of the heart; making them crucial diagnostic tools that provide valuable insights into cardiac pathologies. They are routinely administered dur- ing physical examinations and are often the first test for patients with known or suspected heart disease. Due to the high level of human-expertise required to interpret ECGs, and the volume of ECG recordings, manual real-time interpretation and monitoring of ECG waveforms is impracti- cal. This dissertation addresses the need for automated identification of electrocardiogram (ECG) waveform markers by proposing innovative algorithms that combine machine learning and signal processing techniques.
Unlike rule-based and signal processing algorithms which lack adaptability and struggle with marker heterogeneity, we introduce four sets of algorithms designed for adaptability. These algo- rithms integrate signal processing principles with machine learning paradigms and focus on detect- ing key ECG markers—deviations in R peak recurrence, spatial and temporal attributes of P and T waves, emergence of the U wave, variations in the ST segment, and inversion and elevation of the T wave—that are known to reflect changes in the cardiac condition and aid in diagnosing cardiac arrhythmias. We also explored the application of embeddings from pre-trained transformer-encoder models to detect non-normal sinus rhythm. In a side-by-side comparison, the algorithms we propose demonstrate superior recall, precision, and/or inference time compared to traditional approaches.
The work included in this dissertation offers a comprehensive and automated system for de- tecting medical conditions, allowing for remote patient monitoring and precise diagnosis of cardiac anomalies. The development of these algorithms not only contributes to the field of computational electrophysiology but also holds potential for advancing ECG analysis and advancing cardiovascular healthcare. These novel algorithms stand to facilitate earlier diagnosis, improve automated patient care, and enable timely intervention for cardiovascular diseases, which remain a leading cause of mortality.
Share
Citation/Export: |
|
Social Networking: |
|
Details
Item Type: |
University of Pittsburgh ETD
|
Status: |
Unpublished |
Creators/Authors: |
|
ETD Committee: |
|
Date: |
3 June 2024 |
Date Type: |
Publication |
Defense Date: |
8 February 2024 |
Approval Date: |
3 June 2024 |
Submission Date: |
16 March 2024 |
Access Restriction: |
2 year -- Restrict access to University of Pittsburgh for a period of 2 years. |
Number of Pages: |
141 |
Institution: |
University of Pittsburgh |
Schools and Programs: |
Swanson School of Engineering > Bioengineering |
Degree: |
PhD - Doctor of Philosophy |
Thesis Type: |
Doctoral Dissertation |
Refereed: |
Yes |
Uncontrolled Keywords: |
Machine Learning, Cardiac, Artificial Intelligence, ECG |
Date Deposited: |
03 Jun 2024 14:38 |
Last Modified: |
03 Jun 2024 14:38 |
URI: |
http://d-scholarship.pitt.edu/id/eprint/45875 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |