Skip to main content

View full-text article in PMC

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

Table 3.

Performance of Physicians According to Reader Groups

Reader Groups	Area Under the Receiver Operating Characteristic Curve	Area Under the Alternative Free- response Receiver Operating Characteristic Curve	Sensitivity	Specificity	True Detection Rate
Session 1 (physician reading only)
Nonradiology physicians	0.746 (0.552–0.940)	0.664 (0.466–0.861)	0.723 (0.677–0.765)	0.670 (0.627–0.711)	0.582 (0.543–0.620)
P value^a	.0230	.0088
Board-certified radiologists	0.946 (0.911–0.982)	0.900 (0.856–0.943)	0.906 (0.874–0.932)	0.948 (0.925–0.966)	0.797 (0.764–0.827)
P value^a	.0082	.0003
Thoracic radiologists	0.971 (0.948–0.993)	0.925 (0.890–0.959)	0.952 (0.927–0.970)	0.930 (0.904–0.951)	0.870 (0.842–0.894)
P value^a	0.0218	0.0001
Session 2 (physician reading with DLAD assistance)
Nonradiology physicians	0.850 (0.694–1.005)	0.781 (0.598–0.965)	0.848 (0.810–0.881)	0.800 (0.762–0.834)	0.724 (0.688–0.758)
P value^b	.0610	.0236	<.0001	<.0001	<.0001
Board-certified radiologists	0.961 (0.933–0.988)	0.924 (0.891–0.957)	0.930 (0.901–0.953)	0.954 (0.932–0.971)	0.849 (0.819–0.875)
P value^b	.0606	.0353	.0075	.0833	<.0001
Thoracic radiologists	0.977 (0.957–0.997)	0.942 (0.913–0.971)	0.964 (0.941–0.980)	0.936 (0.911–0.956)	0.897 (0.871–0.919)
P value^b	.1623	.0036	.0587	.2568	.0004

^aComparison of performance with deep learning–based automatic detection (DLAD) algorithm.

^bComparison of performance with session 1.