Deep learning model for pleural effusion detection via active learning and pseudo-labeling: a multisite study

BMC Medical Imaging

Table 1 Basic characteristics for 600 external validation dataset

	Cases (N)	Pleural Effusion ^a		P-value
	Cases (N)	Presence	Absence	P-value
Gender				0.0759
Female	266	116	150
Male	332	169	163
N/A^b	2	1	1
Age Group				< 0.0001
18–49 y/o	171	30	141
50–64 y/o	186	103	83
Above 65 y/o	242	153	89
N/A^b	1	0	1
Data Source				< 0.0001
US	300	160	140
Taiwan	300	126	174
Manufacturer				0.0004
Samsung Electronics	135	64	71
Shimadzu	159	67	92
Toshiba	152	60	92
Others^c	154	95	59
Size of Pleural Effusion				-
Small	196	196	-
Moderate	73	73	-
Large	16	16	-
Size undefined^d	1	1	-
Location of Pleural Effusion				-
Right	134	134	-
Left	85	85	-
Bilateral	62	62	-
Location undefined^d	5	5	-

^a Presence and absence of pleural effusion cases were defined based on the majority agreement between the three radiologists.
^b Cases’ where gender and age were unknown in the dataset.
^c Other X-ray manufacturers include Konica Minolta, GE Healthcare, Drtech, Canon Inc., Siemens, Oehm und Rehbein GmbH, Philips Medical Systems, Swissray, Kodak, Agfa, Fujifilm, and unknown.
^d Cases where only two radiologists agreed on the presence of pleural effusion and the size or location of the pleural effusion was in disagreement between the two radiologists. Thus, these cases were marked as undefine.

ISSN: 1471-2342