Table 2.

Results for the deep learning classifier and human readings of study datasets

COVID-19 positive versus negative
Classifier validation (cross-validation)
	Positive (n)	Negative (n)
Assigned positive	195	46
Assigned negative	55	204
	Sensitivity	Specificity
	0.78* (0.74–0.81)	0.82* (0.78–0.85)
	LR+	LR-
	4.24* (3.24–5.55)	0.27* (0.21–0.34)
Classifier independent testing
	Positive (n)	Negative (n)
Assigned positive	59	7
Assigned negative	15	29
	Sensitivity	Specificity
	0.80* (0.72–0.86)	0.81* (0.73–0.87)
	LR+	LR-
	4.10* (2.09–8.05)	0.25* (0.16–0.41)
Human independent testing (Reader 1)
	Positive (n)	Negative (n)
Assigned positive	47	8
Assigned negative	27	28
	Sensitivity	Specificity
	0.64 (0.52–0.74)	0.78 (0.61–0.90)
	LR+	LR-
	2.86 (1.51–5.39)	0.47 (0.33–0.66)
Human independent testing (Reader 2)
Assigned positive	47	5
Assigned negative	27	31
	Sensitivity	Specificity
	0.64 (0.52–0.74)	0.86 (0.71–0.95)
	LR+	LR-
	4.57 (1.99–10.50)	0.42 (0.31–0.59)

Data are presented as value and 95% confidence interval. *p < 0.005. COVID-19 Coronavirus disease 2019, LR+ Positive likelihood ratio, LR- Negative likelihood ratio