Key result
Single-lead ECG plus PPG machine learning detects LVEF <55% with ~0.91 AUC.
Why the study?
Models for assessing left ventricular systolic function using ECG and photoplethysmography parameters from a single-channel monitor were developed to improve noninvasive evaluation.
Can machine learning models based on single-channel ECG and PPG accurately assess left ventricular systolic function compared to echocardiography?
Observational (n=400)
Can machine learning models based on single-channel ECG and PPG accurately assess left ventricular systolic function compared to echocardiography?
Answers are grounded in this paper and cite their sources.
Machine learning models using single-channel ECG and PPG data can accurately detect reduced left ventricular ejection fraction, offering a potential tool for accessible screening of systolic dysfunction.
Objective. To develop the models for assessing left ventricular (LV) systolic function based on electrocardiogram (ECG) and photoplethysmogram (PPG) parameters recorded by a single-channel ECG monitor with photoplethysmography function. Material and methods. There were 400 patients. All ones underwent echocardiography with analysis of LV ejection fraction (EF) and velocity time integral (VTI). After that, registration of ECG and PPG was performed using a single-channel ECG scanner with function of photoplethysmography. Then all data were transmitted to a single server, where ECG and PPG parameters were calculated. ECG- and PPG-based models for assessing LV systolic function were built using Lasso regression and random forest classifier (RFC). Results. Models were obtained for EF <55%, 40%, 30% and VTI <16 and 13 cm, respectively. For each model, we analyzed area under ROC curve (AUC), sensitivity and specificity. The following data were obtained using Lasso regression: for EF<55% AUC 0.857 (sensitivity 0.818, specificity 0.860); for EF<40% AUC 0.971; for EF<30% AUC 0.982; for VTI <13 AUC 0.754, for VTI <16 AUC 0.746. RFC revealed the following data: AUC for EF <55% was 0.913; for EF <40% — 0.955; for EF <30% — 0.962; for VTI <13 AUC was 0.776, for VTI <16 — 0.782. Conclusion. ECG- and PPG-based models showed high accuracy in assessment of LV systolic function.
No takes yet. Share an insight, caveat, or question.
Sagirova et al. (2023) conducted an observational in Left ventricular systolic dysfunction (n=400). Single-channel ECG monitor with photoplethysmography vs. Echocardiography was evaluated on Detection of left ventricular ejection fraction <55%. Machine learning models using single-channel ECG and photoplethysmography accurately detected left ventricular ejection fraction <55% with an AUC of 0.913 using a random forest classifier.
Synapse has enriched 10 related papers. Consider them for comparative context: