Table 2.

Performance metrics (%) for lung region and COVID-19 infected region segmentation 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.

Task	Model	Encoder	Accuracy	IoU	DSC
Lung Segmentation	U-Net	ResNet18	99.07 ± 0.23	95.91 ± 0.47	97.88 ± 0.34
		ResNet50	99.08 ± 0.23	95.93 ± 0.47	97.89 ± 0.34
		DenseNet121	99.1 ± 0.22	96.06 ± 0.46	97.96 ± 0.34
		DenseNet161	99.1 ± 0.22	96.02 ± 0.47	97.94 ± 0.34
		InceptionV4	99.07 ± 0.23	95.9 ± 0.47	97.88 ± 0.34
	U-Net ++	ResNet18	99.07 ± 0.23	95.9 ± 0.47	97.88 ± 0.34
		ResNet50	99.1 ± 0.22	96.04 ± 0.46	97.95 ± 0.34
		DenseNet121	99.11 ± 0.22	96.1 ± 0.46	97.98 ± 0.33
		DenseNet161	99.09 ± 0.23	95.98 ± 0.47	97.92 ± 0.34
		InceptionV4	99.08 ± 0.23	95.96 ± 0.47	97.91 ± 0.34
	FPN	ResNet18	99.06 ± 0.23	95.86 ± 0.47	97.86 ± 0.34
		ResNet50	99.07 ± 0.23	95.91 ± 0.47	97.88 ± 0.34
		DenseNet121	99.12 ± 0.22	96.11 ± 0.46	97.99 ± 0.33
		DenseNet161	99.09 ± 0.23	96.01 ± 0.47	97.94 ± 0.34
		InceptionV4	99.07 ± 0.23	95.92 ± 0.47	97.89 ± 0.34
Infection Segmentation	U-Net	ResNet18	98.02 ± 0.8	82.92 ± 2.16	88.1 ± 1.86
		ResNet50	97.84 ± 0.83	81.73 ± 2.22	87.02 ± 1.93
		DenseNet121	97.98 ± 0.81	82.53 ± 2.18	87.74 ± 1.88
		DenseNet161	97.86 ± 0.83	81.95 ± 2.21	87.19 ± 1.92
		InceptionV4	97.98 ± 0.81	82.03 ± 2.2	87.11 ± 1.92
	U-Net ++	ResNet18	97.9 ± 0.82	82.9 ± 2.16	88.06 ± 1.86
		ResNet50	97.93 ± 0.82	82.59 ± 2.18	87.78 ± 1.88
		DenseNet121	97.97 ± 0.81	83.05 ± 2.15	88.21 ± 1.85
		DenseNet161	97.95 ± 0.81	81.55 ± 2.23	86.66 ± 1.95
		InceptionV4	97.9 ± 0.82	81.13 ± 2.25	86.22 ± 1.98
	FPN	ResNet18	97.84 ± 0.83	81.9 ± 2.21	87.25 ± 1.91
		ResNet50	97.84 ± 0.83	80.83 ± 2.26	86.25 ± 1.98
		DenseNet121	97.99 ± 0.81	82.55 ± 2.18	87.71 ± 1.88
		DenseNet161	97.95 ± 0.81	81.89 ± 2.21	87.08 ± 1.93
		InceptionV4	97.99 ± 0.81	83.08 ± 2.15	88.13 ± 1.86