Table 2.

The difference between various deep learning models.

Models	Groups	AUC (95%CI)	Accuracy	Sensitivity	Specificity
Resnet18	Training	0.980 (0.969-0.991)	0.937	0.956	0.930
	Internal validation	0.981 (0.963-0.998)	0.924	0.925	0.948
	External validation	0.935 (0.888-0.983)	0.839	0.957	0.814
Resnet34	Training	0.979 (0.970-0.989)	0.916	0.995	0.873
	Internal validation	0.974 (0.954-0.995)	0.916	0.925	0.935
	External validation	0.877 (0.808-0.946)	0.796	1.000	0.710
Resnet50	Training	0.988 (0.981-0.994)	0.941	0.961	0.924
	Internal validation	0.977 (0.946-1.000)	0.962	0.962	0.961
	External validation	0.939 (0.894-0.984)	0.860	1.000	0.800
Resnet101	Training	0.975 (0.960-0.990)	0.946	0.966	0.949
	Internal validation	0.992 (0.984-1.000)	0.946	1.000	0.897
	External validation	0.968 (0.935-1.000)	0.914	1.000	0.929
Resnet152	Training	0.982 (0.971-0.992)	0.946	0.928	0.958
	Internal validation	0.981 (0.963-0.999)	0.931	0.944	0.948
	External validation	0.950 (0.911-0.990)	0.828	1.000	0.857
Densenet121	Training	0.996 (0.994-0.999)	0.965	0.995	0.952
	Internal validation	0.985 (0.963-1.000)	0.962	0.925	1.000
	External validation	0.903 (0.837-0.969)	0.806	0.913	0.829
Densenet201	Training	0.995 (0.992-0.998)	0.967	0.966	0.971
	Internal validation	0.985 (0.963-1.000)	0.969	0.962	0.974
	External validation	0.953 (0.914-0.993)	0.860	0.957	0.829
Inception v3	Training	0.986 (0.977-0.995)	0.944	0.961	0.949
	Internal validation	0.987 (0.973-1.000)	0.946	0.944	0.961
	External validation	0.929 (0.873-0.985)	0.839	0.826	0.929

AUC, area under the receiver operating characteristic curve; 95%CI, 95% confidence intervals.