Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

In Search of an Optimal Subset of Electrocardiogram Features to Augment the Diagnosis of Acute Coronary Syndrome at the Emergency Department

Bouzid, Zeineb (2020) In Search of an Optimal Subset of Electrocardiogram Features to Augment the Diagnosis of Acute Coronary Syndrome at the Emergency Department. Master's Thesis, University of Pittsburgh. (Unpublished)

This is the latest version of this item.

Primary Text

Download (3MB) | Preview


The electrophysiology of acute myocardial ischemia is well understood; yet clinical practice primarily relies on classical ST amplitude measures. This translates into poor diagnostic sensitivity for identifying acute coronary syndrome (ACS). Machine learning could help identify an optimal subset of features to augment clinicians' decision during patient evaluation. We
sought to compare the accuracy of supervised classifiers using electrocardiogram (ECG) feature subsets selected based on data-driven techniques or domain-specific knowledge.
This was an observational study of two prospective cohorts of consecutive patients evaluated at the emergency department for suspected ACS (Cohort 1: n=745, age 59±17, 42% Female; Cohort 2: n=499, age 59±16, 49% Female). A total of 554 temporal-spatial waveform
features were extracted from baseline 12-lead ECGs using manufacturer-specific software. We used multiple algorithms to identify a subset of 229 data-driven features. Additionally, we selected a subset of 65 physiology-driven features that are mechanistically linked to myocardial ischemia. Using these two subsets of features, we evaluated logistic regression (LR) and artificial neural network (ANN) classifiers using 10-fold cross-validation on cohort 1 with independent testing on cohort 2. Our results show that classifiers with data-driven features were superior during model training (Area under the ROC curve: 0.81±0.06 vs 0.76±0.09 for LR, and 0.85±0.07 vs. 0.80±0.05 for ANN), but they generalized poorly to testing data (Area under the ROC curve: 0.68 vs 0.76 for LR, and 0.72 vs. 0.77 for ANN). In addition
to classical ST and T wave amplitudes, the following features were found to be important in ACS classification: T peak-Tend interval; QRS and T axes with corresponding angles; T loop morphology, and principal component analysis ratio of ECG waveforms.
In this study, we identified a subset of novel ECG features that would improve ACS detection. These features guided by domain-specific knowledge yielded stable LR classifiers highly adaptable to be implemented in clinical decision support tools.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Bouzid, Zeinebzeb12@pitt.eduzeb12
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairSejdic, Ervinesejdic@pitt.eduesejdic
Committee MemberAkcakaya, Muratakcakaya@pitt.eduakcakaya
Committee MemberDallal, AhmedAHD12@pitt.eduAHD12
Committee MemberAl-Zaiti, SalahSSA33@pitt.eduSSA33
Thesis AdvisorSejdic, Ervinesejdic@pitt.eduesejdic
Date: 27 September 2020
Date Type: Publication
Defense Date: 14 July 2020
Approval Date: 27 September 2020
Submission Date: 22 July 2020
Access Restriction: 1 year -- Restrict access to University of Pittsburgh for a period of 1 year.
Number of Pages: 74
Institution: University of Pittsburgh
Schools and Programs: Swanson School of Engineering > Electrical and Computer Engineering
Degree: MS - Master of Science
Thesis Type: Master's Thesis
Refereed: Yes
Uncontrolled Keywords: machine learning, dimensionality reduction, acute coronary syndrome, electrocardiogram, ischemia.
Date Deposited: 27 Sep 2020 22:10
Last Modified: 27 Sep 2021 05:15

Available Versions of this Item


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item