Table 1. Dataset for model development and evaluation.

	Dataset for model development		Test dataset
CXR images according to data source (normal vs. pneumonia)	n = 157,016 (120,722 vs. 36,294) from NIH (60,361 vs. 36,294) from KNTA (60,361 vs. 0)		normal: NIH(n = 106), KNTA (n = 106) pneumonia from GUGMC (n = 212)
	Training dataset (70%)	Validation dataset (30%)
	n = 109,912 (84,506 vs. 25,406)	n = 47,104 (36,216 vs. 10,888)	n = 424 (212 vs. 212)
Number of patients (male vs. female)	62,703 (32,065 vs. 30,638)	28,463 (14,746 vs. 13,717)	424 (288 vs. 136)
Age (mean ± SD)	47 ± 16	47 ± 16	54 ± 13

Abbreviations: CXR, chest X-ray; NIH, National Institutes of Health; KNTA, Korean National Tuberculosis Association; GUGMC, Gachon University Gil Medical Center.