Related Experiment Videos
ECG-Based Classification of Resuscitation Cardiac Rhythms for Retrospective Data Analysis
IEEE Transactions on Bio-Medical Engineering
|April 4, 2017
Summary
This study developed an artificial neural network (ANN) algorithm to automatically classify cardiac rhythms during resuscitation, achieving 78.5% accuracy. The system shows promise for efficient rhythm monitoring, though accuracy for organized rhythms requires improvement.
Area of Science:
- Cardiology
- Biomedical Engineering
- Artificial Intelligence in Medicine
Background:
- Monitoring heart rhythm during resuscitation is crucial for improving treatment quality.
- Manual annotation of resuscitation rhythms (ventricular tachycardia, ventricular fibrillation, pulseless electrical activity, asystole, pulse-generating rhythm) is time-consuming and impractical for large datasets.
Purpose of the Study:
- To develop and evaluate ECG-based algorithms for the automatic classification of cardiac rhythms during resuscitation.
Main Methods:
- A dataset of 1631 3-s ECG segments from 298 out-of-hospital cardiac arrest patients was analyzed.
- 47 wavelet- and time-domain features were extracted and selected using a wrapper-based approach.
- Classifiers including Bayesian decision theory, k-nearest neighbor, ANN, and ensemble of decision trees were evaluated.
Main Results:
- The best performing algorithm was an ANN with Bayesian regularization, utilizing 14 features.
- The proposed algorithm achieved an overall accuracy of 78.5%.
- Sensitivities and positive predictive values varied by rhythm, with highest for Asystole (88.7%, 91.0%) and Ventricular Fibrillation (86.2%, 83.8%).
Conclusions:
- Automatic classification of resuscitation cardiac rhythms is feasible using ECG-based algorithms.
- The developed ANN algorithm represents a step towards more efficient rhythm classification with minimal expert feedback.
- Accuracy for organized rhythms like pulseless electrical activity and pulse-generating rhythm remains a challenge.