Table 2.

COVID-19 detection performance results on Pneumonia vs. COVID-19 test set when trained on real data; real + synthetic data; and only synthetic data (i.e., only ${\mathcal{G}}_{\mathit{PC}}$ is used for positive class examples in training each model)

Model	Real			Real + ${\mathcal{G}}_{\mathit{PC}}$			Real + ${\mathcal{G}}_{\mathit{PC}}$ + ${\mathcal{G}}_{\mathit{NC}}$			Only Synthetic
	SEN (%) $\uparrow$	FN $\downarrow$	Skew$\downarrow$	SEN (%) $\uparrow$	FN $\downarrow$	Skew$\downarrow$	SEN(%) $\uparrow$	FN $\downarrow$	Skew$\downarrow$	SEN (%) $\uparrow$	FN $\downarrow$
VGG-16	8.69	42	22.9	29.50	24	0.95	52.17	22	0.19	39.13	28
ResNet-50	21.73	36	22.9	36.95	29	0.95	41.30	27	0.19	13.04	40
DenseNet-102	4.34	44	22.9	21.74	36	0.95	32.43	32	0.19	6.52	43
DenseNet-121 + BGT	32.60	31	22.9	41.30	27	0.95	47.82	24	0.19	32.60	31

Boldface numbers indicate the best performance