Machine Learning Methods for Robust ECG Beat Detection – UROP Spring Symposium 2021

Machine Learning Methods for Robust ECG Beat Detection

Anastacia Gusikhin

UROP Logo

Pronouns: She/Her

Research Mentor(s): Mohammed Saeed, Faculty Member / Clinical Lecturer
Research Mentor School/College/Department: Department of Internal Medicine, Michigan Medicine
Presentation Date: Thursday, April 22, 2021
Session: Session 4 (2pm-2:50pm)
Breakout Room: Room 17
Presenter: 6

Event Link

Abstract

This research was conducted to determine what types of machine learning algorithms were best to determine whether an individual is experiencing heart arrhythmias—such as atrial fibrillation, a type of arrhythmia that can lead to a number of fatal conditions such as blood clots, stroke, and heart failure. Using Python and the sklearn library, a number of machine learning models were tested for accuracy of ECG peak detection, which included Logistic Regression, Linear Discriminant Analysis, K-Nearest Neighbors, Classification and Regression Trees, Gaussian Naive Bayes, and Support Vector Machines. The accuracies varied based on the proportion of noise within the ECG file, so an introspective algorithm was developed to choose the optimal peak detection algorithm based on the estimated noise level measured in an ECG. This can be used in the future as a convenient way to accurately determine the presence of heart arrhythmias using wearable devices such as a smart watch.

Authors: Mohammed Saeed, Anastacia Gusikhin
Research Method: Computer Programming

lsa logoum logo