TABLE 5. Quantitative Evaluation Results of Different Modules Used in Our Proposed Model. RB: Residual Block, RRB: Residual Refinement Block, PCB: Pyramid Convolution Block, CAB: Channel Attention Block.

Method	Infection type	DSC	Sen.	Spec.
(1) ResNet-101+RB (Backbone)	Ground-Glass Opacities	0.7034 ± 0.044	0.8011 ± 0.019	0.9714 ± 0.011
	Interstitial Infiltrates	0.6958 ± 0.026	0.7969 ± 0.020	0.9799 ± 0.009
	Consolidation	0.7934 ± 0.020	0.8289 ± 0.025	0.9806 ± 0.012
(2) ResNet-101+RRB	Ground-Glass Opacities	0.7148 ± 0.041	0.8246 ± 0.021	0.9729 ± 0.007
	Interstitial Infiltrates	0.7198 ± 0.021	0.8101 ± 0.018	0.9832 ± 0.010
	Consolidation	0.8066 ± 0.020	0.8536 ± 0.018	0.9833 ± 0.009
(3) ResNet-101+RRB+CAB	Ground-Glass Opacities	0.7278 ± 0.039	0.8402 ± 0.019	0.9781 ± 0.006
	Interstitial Infiltrates	0.7293 ± 0.026	0.8145 ± 0.0201	0.9862 ± 0.008
	Consolidation	0.8605 ± 0.0151	0.8502 ± 0.019	0.9891 ± 0.005
(4) ResNet-101+RRB+CAB+PCB	Ground-Glass Opacities	0.7422 ± 0.038	0.8593 ± 0.018	0.9742 ± 0.005
	Interstitial Infiltrates	0.7384 ± 0.021	0.8268 ± 0.020	0.9869 ± 0.005
	Consolidation	0.8769 ± 0.015	0.8645 ± 0.017	0.9889 ± 0.007