Table 2.

Detection of stressed mothers by self-supervised learning trained on the FELICITy and public datasets.

Source	Accuracy	F1 Score	Sensitivity	Specificity	PPV	NPV	AUROC
FELICITy dataset
aECG	0.795 ± 0.023	0.777 ± 0.022	0.779 ± 0.031	0.809 ± 0.045	0.777 ± 0.039	0.812 ± 0.020	0.794 ± 0.022
mECG	0.931 ± 0.093	0.925 ± 0.101	0.924 ± 0.101	0.937 ± 0.087	0.926 ± 0.102	0.936 ± 0.086	0.931 ± 0.094
Public datasets
aECG	0.936 ± 0.002	0.930 ± 0.003${}^{\ast }$	0.926 ± 0.008${}^{\ast }$	0.945 ± 0.004${}^{\ast }$	0.935 ± 0.004${}^{\ast }$	0.938 ± 0.006${}^{\ast }$	0.936 ± 0.002${}^{\ast }$
mECG	0.982 ± 0.003	0.980 ± 0.003${}^{\#\ast }$	0.982 ± 0.004${}^{\#\ast }$	0.982 ± 0.006${}^{\#\ast }$	0.979 ± 0.007${}^{\#}$	0.985 ± 0.003${}^{\#}$	0.982 ± 0.002${}^{\#\ast }$

${}^{\ast }$Public versus FELICITy dataset, Mann Whitney U test.

${}^{\#}$maternal ECG (mECG) versus abdominal ECG (aECG) within the same dataset, Mann Whitney U test.

Statistical significance at p < 0.025 accounting for two comparisons (using Bonferroni–Holm correction).

PPV positive predictive values, NPV negative predictive values, AUROC area under the receiver operating characteristic.