Table 2.

A comparison of our model's overall F1 scores between using the optimal ECG-lead subset and using the complete 12-lead ECG on different datasets. p-values were calculated using the independent two-sample t-test between the overall F1 scores using the optimal 4-lead ECG and using the complete 12-lead ECG. A p-value < 0.05 means the optimal 4-lead ECG system provides a significant improvement in classifying ECG abnormalities over the complete 12-lead ECG.

datasets		F1 scores using the optimal 4-lead ECG	F1 scores using the complete 12-lead ECG	p-value
CPSC 2018 (8 ECG abnormality types)	validation	0.802 ± 0.004	0.789 ± 0.004	0.000
	held-out test	0.764 ± 0.005	0.769 ± 0.005	0.123
PhysioNet/CinC 2020 (4 ECG abnormality types)	external test	0.547 ± 0.001	0.537 ± 0.001	0.000