. 2018 Nov 8;69(5):739–747. doi: 10.1093/cid/ciy967

Table 2.

Performance of Deep Learning–based Automatic Detection Algorithm in the 6 External Validation Datasets

Performance Measures	Seoul National University Hospital Dataset	Boramae Medical Center Dataset	Kyunghee University Hospital at Gangdong Dataset	Daejeon Eulji Medical Center Dataset	Montgomery Dataset	Shenzhen Dataset
Area under the receiver operating characteristic curve	0.993 (0.984–1.002)	0.979 (0.954–1.005)	1.000 (0.999–1.000)	1.000 (0.999–1.000)	0.996 (0.991–1.001)	0.977 (0.967–0.987)
Area under the alternative free-response receiver operating characteristic curve	0.993 (0.983–1.003)	0.981 (0.960–1.001)	0.994 (0.987–1.001)	1.000 (0.999–1.000)	0.996 (0.990–1.002)	0.973 (0.963–0.984)
Sensitivity_SEN^a	0.952 (0.881–0.987)	0.943 (0.860–0.984)	1.000 (0.965–1.000)	1.000 (0.949–1.000)	1.000 (0.932–1.000)	0.947 (0.916–0.969)
Specificity_SEN^a	1.000 (0.964–1.000)	0.957 (0.880–0.991)	0.914 (0.823–0.968)	0.980 (0.930–0.998)	0.938 (0.860–0.979)	0.911 (0.875–0.940)
True detection rate_SEN^a	0.962 (0.914–0.988)	0.945 (0.894–0.976)	1.000 (0.981–1.000)	1.000 (0.984–1.000)	1.000 (0.956–1.000)	0.953 (0.931–0.970)
Sensitivity_SPE^a	0.843 (0.747–0.914)	0.900 (0.805–0.959)	0.990 (0.947–1.000)	0.986 (0.923–1.000)	0.846 (0.719–0.931)	0.841 (0.796–0.879)
Specificity_SPE^a	1.000 (0.964–1.000)	1.000 (0.949–1.000)	1.000 (0.949–1.000)	1.000 (0.964–1.000)	1.000 (0.955–1.000)	0.991 (0.973–0.998)
True detection rate_SPE^a	0.750 (0.667–0.821)	0.759 (0.681–0.826)	0.806 (0.743–0.860)	0.719 (0.656–0.776)	0.719 (0.609–0.813)	0.771 (0.731–0.807)

^aSubscript SEN indicates the high-sensitivity cutoff; subscript SPE indicates the high-specificity cutoff.