Table 3.

COVID-19 detection performance results (%) computed over test (unseen) set with three network models, and five encoder architectures. x ± y means that the achieved metric value is x with standard deviation y.

Model	Encoder	Accuracy	Precision	Sensitivity	F1-score	Specificity
U-Net	ResNet18	98.89 ± 0.6	99.14 ± 0.53	98.63 ± 0.67	98.88 ± 0.6	99.14 ± 0.53
	ResNet50	98.89 ± 0.6	98.47 ± 0.7	99.31 ± 0.48	98.89 ± 0.6	98.46 ± 0.71
	DenseNet121	98.8 ± 0.62	97.98 ± 0.81	99.66 ± 0.33	98.81 ± 0.62	97.94 ± 0.82
	DenseNet161	98.71 ± 0.65	97.97 ± 0.81	99.49 ± 0.41	98.72 ± 0.65	97.94 ± 0.82
	InceptionV4	98.03 ± 0.8	98.28 ± 0.75	97.77 ± 0.85	98.02 ± 0.8	98.28 ± 0.75
U-Net ++	ResNet18	99.23 ± 0.5	100 ± 0	98.46 ± 0.71	99.22 ± 0.5	100 ± 0
	ResNet50	99.14 ± 0.53	99.83 ± 0.24	98.46 ± 0.71	99.14 ± 0.53	99.83 ± 0.24
	DenseNet121	99.23 ± 0.5	99.14 ± 0.53	99.31 ± 0.48	99.22 ± 0.5	99.14 ± 0.53
	DenseNet161	98.2 ± 0.76	97.95 ± 0.81	98.46 ± 0.71	98.2 ± 0.76	97.94 ± 0.82
	InceptionV4	98.2 ± 0.76	98.45 ± 0.71	97.94 ± 0.82	98.19 ± 0.77	98.46 ± 0.71
FPN	ResNet18	98.54 ± 0.69	97.48 ± 0.9	99.66 ± 0.33	98.56 ± 0.68	97.43 ± 0.91
	ResNet50	98.46 ± 0.71	98.46 ± 0.71	98.46 ± 0.71	98.46 ± 0.71	98.46 ± 0.71
	DenseNet121	98.97 ± 0.58	99.65 ± 0.34	98.28 ± 0.75	98.96 ± 0.58	99.66 ± 0.33
	DenseNet161	98.11 ± 0.78	97.3 ± 0.93	98.97 ± 0.58	98.13 ± 0.78	97.26 ± 0.94
	InceptionV4	99.23 ± 0.5	99.31 ± 0.48	99.14 ± 0.53	99.22 ± 0.5	99.31 ± 0.48