Table 2.

Comparison results of our method vs. state-of-the-art methods performed on the CC-CCII dataset.

Annotation	Method	Patient Acc. (%)	Precision (%)	Sensitivity (%)	Specificity (%)	AUC (%)
Patient-level	ResNet-50 [165]	53.44	64.45	63.03	35.71	53.24
	COVID-Net [177]	57.13	62.53	84.70	6.16	49.58
	COVNet [178]	69.96	70.20	93.33	26.75	81.61
	VB-Net [179]	76.11	75.84	92.73	45.38	88.34
	Ours	89.97	92.99	91.44	87.25	95.53
Image-level	ResNet-50 [165]	52.56	61.60	71.27	18.06	50.19
	COVID-Net [177]	60.03	64.81	83.91	15.98	58.39
	COVNet [178]	75.55	79.90	83.24	61.37	79.48
	Ours	80.41	88.56	80.15	80.89	86.06