Deep learning of ECG waveforms for diagnosis of heart failure with a reduced left ventricular ejection fraction

Figure 1. Study flow chart—Among the patients hospitalized with acute heart failure, subjects with no matching echocardiographic results within 1 months and electrocardiograms other than closest match to the echocardiographic results were excluded. ECG electrocardiogram.

Figure 2. (a) Proportion of DeepECG-HFrEF (+) according to actual EF (b) Distribution of correct and incorrect cases of DeepECG-HFrEF according to actual EF—The proportion of patients diagnosed with DeepECG-HFrEF (+) increased with the lower actual EF. The false-positives and false-negatives of DeepECG-HFrEF were more likely yielded when the EF was near 40%. ECG electrocardiogram, EF ejection fraction.

Figure 3. Scatter plot demonstrating observed EF and DeepECG-HFrEF prediction—The proportion of true-positives was higher with lower actual EF even in the scatter plot. EF ejection fraction.

Figure 4. Forest plot depicting AUC values (95% confidence interval) of the DeepECG-HFrEF algorithm in identifying LVSD in different subpopulation. The forest plot of DeepECG-HFrEF algorithm showed similar performance among various clinical parameters. AUC area under the receiver-operator characteristic curve, BNP B-type natriuretic peptide, EF ejection fraction, eGFR estimated glomerular filtration rate, HF heart failure, LAD left axis deviation, RAD right axis deviation.

Figure 5. (a) Kaplan–Meier curve for mortality at 5-year follow up according to the DeepECG-HFrEF (Total ECGs = 1291) (b) Kaplan–Meier curve for mortality at 5-year follow up according to the DeepECG-HFrEF among patients with actual EF < 40%—The patients classified as DeepECG-HFrEF positive showed worse 5-year survival. ECG electrocardiogram; EF ejection fraction.