Prediction of significant congenital heart disease in infants and children using continuous wavelet transform and deep convolutional neural network with 12-lead electrocardiogram

Lee, Yu-Shin; Chung, Hung-Tao; Lin, Jainn-Jim; Hwang, Mao-Sheng; Liu, Hao-Chuan; Hsu, Hsin-Mao; Chang, Ya-Ting; Peng, Syu-Jyun

doi:10.1186/s12887-025-05628-2

Table 5 Impact of significance in prediction of CHD: results of transfer learning using various pre-trained models

From: Prediction of significant congenital heart disease in infants and children using continuous wavelet transform and deep convolutional neural network with 12-lead electrocardiogram

NOR vs RHA + RHB
	ResNet- 18	InceptionResNet-V2	NASNetMobile
Sensitivity	0.719 ± 0.029 (0.683–0.755)	0.804 ± 0.049 (0.742–0.865)	0.750 ± 0.041 (0.699–0.800)
Specificity	0.633 ± 0.044 (0.579–0.688)	0.551 ± 0.051 (0.488–0.615)	0.609 ± 0.078 (0.512–0.706)
Accuracy	0.686 ± 0.033 (0.645–0.727)	0.707 ± 0.041 (0.656–0.758)	0.696 ± 0.012 (0.681–0.711)
F1 score	0.739 ± 0.028 (0.704–0.773)	0.772 ± 0.035 (0.728–0.816)	0.752 ± 0.011 (0.738–0.767)
AUROC	0.738 ± 0.029 (0.702–0.774)	0.758 ± 0.027 (0.725–0.791)	0.757 ± 0.022 (0.730–0.785)
NOR + RHA vs RHB
	ResNet- 18	InceptionResNet-V2	NASNetMobile
Sensitivity	0.735 ± 0.032 (0.695–0.775)	0.759 ± 0.052 (0.695–0.824)	0.747 ± 0.074 (0.655–0.839)
Specificity	0.839 ± 0.022 (0.812–0.865)	0.803 ± 0.055 (0.734–0.872)	0.824 ± 0.043 (0.770–0.878)
Accuracy	0.789 ± 0.009 (0.778–0.799)	0.782 ± 0.036 (0.737–0.827)	0.787 ± 0.019 (0.764–0.811)
F1 score	0.770 ± 0.014 (0.753–0.787)	0.770 ± 0.040 (0.721–0.819)	0.770 ± 0.034 (0.728–0.811)
AUROC	0.852 ± 0.011 (0.838–0.866)	0.867 ± 0.020 (0.842–0.892)	0.865 ± 0.020 (0.840–0.890)
NOR vs LHA + LHB
	ResNet- 18	InceptionResNet-V2	NASNetMobile
Sensitivity	0.617 ± 0.037 (0.571–0.663)	0.640 ± 0.018 (0.618–0.661)	0.547 ± 0.037 (0.501–0.592)
Specificity	0.857 ± 0.044 (0.803–0.911)	0.834 ± 0.045 (0.778–0.890)	0.895 ± 0.031 (0.856–0.934)
Accuracy	0.704 ± 0.010 (0.691–0.717)	0.710 ± 0.011 (0.697–0.724)	0.673 ± 0.017 (0.652–0.695)
F1 score	0.726 ± 0.017 (0.704–0.747)	0.737 ± 0.008 (0.727–0.747)	0.680 ± 0.025 (0.649–0.711)
AUROC	0.817 ± 0.021 (0.790–0.843)	0.802 ± 0.019 (0.777–0.826)	0.786 ± 0.025 (0.754–0.817)
NOR + LHA vs LHB
	ResNet- 18	InceptionResNet-V2	NASNetMobile
Sensitivity	0.626 ± 0.060 (0.551–0.701)	0.696 ± 0.038 (0.649–0.744)	0.481 ± 0.022 (0.454–0.509)
Specificity	0.815 ± 0.014 (0.797–0.832)	0.779 ± 0.043 (0.725–0.832)	0.912 ± 0.013 (0.895–0.929)
Accuracy	0.736 ± 0.033 (0.695–0.777)	0.744 ± 0.033 (0.704–0.785)	0.732 ± 0.008 (0.721–0.742)
F1 score	0.664 ± 0.050 (0.602–0.725)	0.695 ± 0.035 (0.652–0.738)	0.600 ± 0.016 (0.580–0.620)
AUROC	0.804 ± 0.031 (0.765–0.842)	0.816 ± 0.035 (0.773–0.859)	0.814 ± 0.017 (0.793–0.835)

Mean ± SD (95% CI)
SD standard deviation, CI confidence interval

Back to article page

ISSN: 1471-2431

Contact us

General enquiries: journalsubmissions@springernature.com

BMC Pediatrics

Contact us