Table 2. Selected applications of ML algorithms applied to the ECG.

Application	ML model	Training dataset (patients)	Testing dataset (patients)	AUC	Sensitivity (%)	Specificity (%)	Accuracy (%)	Publication (Ref. #)
Screening hyperkalemia from a 2-lead ECG in patients with CKD	CNN	449,380	61,965	0.88	90.2	63.2	–	Galloway et al. (16)
Detecting asymptomatic LV dysfunction from a 12-lead ECG	CNN	44,959	52,870	0.93	86.3	85.7	85.7	Attia et al. (17)
Predicting AF in asymptomatic patients in sinus rhythm from a 12-lead ECG	CNN	126,526	54,396	0.90	82.3	83.4	83.3	Attia et al. (18)
Detecting LV hypertrophy from a 12-lead ECG	CNN	12,648	5,476	0.87	61.3	89.6	85.1	Kwon et al. (19)
Predicting gender & age from a 12-lead ECG	CNN	499,727	275,056	0.94	87.8	86.8	87.0	Attia et al. (20)
Diagnosing arrhythmia from a single lead ECG	CNN	29,163	328	0.97	78.0	–	80.9	Rajpurkar et al. (21)
Detecting MI from a 12-lead ECG	CNN	–	290	–	93.3	89.7	–	Strodthoff et al. (22)

ML, machine learning; CNN, convolutional neural network; AUC, area under the curve; AF, atrial fibrillation; CKD, chronic kidney disease; LV, left ventricular; MI, myocardial infarction; ECG, electrocardiogram.