TABLE 3.

The DAR and kappa between AI software, every grader (A,B,C) and gold standard for detecting any DR and referable DR.

	Any DR detection(95% CI)		Referable DR detection(95% CI)
	DAR	Kappa	DAR	Kappa
Grader A	92.19% (90.44% ∼ 93.72%)	0.930 (0.915 ∼ 0.944)	97.06% (95.88% ∼ 97.98%)	0.940 (0.927 ∼ 0.953)
Grader B	88.61% (86.58% ∼ 90.44%)	0.927 (0.912 ∼ 0.942)	96.32% (95.03% ∼ 97.36%)	0.929 (0.915 ∼ 0.943)
Grader C	80.53% (78.05% ∼ 82.85%)	0.804 (0.779 ∼ 0.828)	92.47% (90.74% ∼ 93.97%)	0.844 (0.823 ∼ 0.865)
AI group	89.30% (87.30% ∼ 91.00%)	0.761 (0.734 ∼ 0.787)	93.11% (91.44% ∼ 94.54%)	0.860 (0.827 ∼ 0.890)

AI = artificial intelligence, CI = confidence interval, DAR = diagnostic accordance rate, DR = diabetic retinopathy.