Share on Facebook
Share on Twitter
Share on LinkedIn
Share by Email
Print
Machine learning may be useful for detecting inaccurate electrocardiogram (ECG) lead placement, according to a systematic review and meta-analysis published in the Journal of Electrocardiology.
Researchers from Ulster University in the United Kingdom searched literature databases through September 2019 for articles in which detection of ECG lead misplacement using machine learning approaches was examined. A final set of 14 articles were included in this study. The studies were conducted in the United States (n=4), Sweden (n=3), Bulgaria and Switzerland (n=3), the United Kingdom (n=2), and the Netherlands (n=2).
The included studies used the following machine learning approaches: decision trees (DT; n=5), artificial neural networks (ANN; n=3), correlation (n=3), support vector machines (SVM; n=3), haisty (n=1), and amplitude threshold (AT; n=1). Half of the studies focused on limb and chest lead interchanges (n=7), 4 studies focused on limb leads, 2 on chest leads, and 1 on chest lead interchanges only.
Studies were evaluated to answer 4 main questions. Why is detecting electrode misplacement important? Which electrodes are most sensitive to displacement? How are misplacements detected? Which machine learning approach is superior for detecting a displacement?
Lead misplacement is important as it may simulate cardiovascular abnormalities including myocardial infarction, ectopic rhythm, or chamber enlargement, leading to incorrect diagnoses. The most frequently misplaced electrodes were V1 and V2 (>50%) and leads V2, V3, V4, and V1 were the most sensitive to displacement.
Machine learning approaches were found to have high sensitivities (DT: 17.9%-99.3%; ANN: 44.5%-99.9%; correlation: 87.0%-97.8%; SVM:56.5%-93.8%; haisty: 84.2%; AT: 20.0%-90.0%) and specificities (DT: 86.6%-100%; ANN:99.8%-99.9%; correlation: 91.0%; SVM:86.6%-99.9%; haisty: 99.9%; AT: ~99.8%) for detecting misplacements.
Chest leads tended to have higher sensitivities (sensitivity: 44.5%-99.95%; specificity: 91%-100%) and limb leads tended to have higher specificities (sensitivity: 20%-99.3%; specificity: 95%-100%).
The algorithm which best detected electrode misplacement was the decision tree, which was successful 5 out of 10 times. The other methods were correct fewer than 4 out of 10 times. The most challenging lead interchange to detect was LA-LL.
This analysis was limited by the examination of multiple machine learning approaches and discrepancies in the leads examined.
“Our findings highlight opportunities for enhancing ECG data quality and decision making through the accurate detection of lead misplacement,” noted the study authors.
Reference
Rjoob K, Bond R, Finlay D, et al. Machine learning techniques for detecting electrode misplacement and interchanges when recording ECGs: A systematic review and meta-analysis. J Electrocardiol. 2020;62:116-123. doi:10.1016/j.jelectrocard.2020.08.013
