Table 5.

Comparison of various methods, including the proposed network (CVR-Net), where the methods are trained on the same dataset and evaluated using an independent test set, not used during training. The top three performing metrics are denoted by bold-font, underline, and double-underline.

Methods	Parameters	CXR-Independent-CL2			CT-Independent-CL2
		Recall	Precision	Accuracy	Recall	Precision	Accuracy
VGG-19	$46$M	0.833	0.846	0.833	0.785	0.816	0.785
Xception	$124$M	0.869	0.881	0.869	0.718	0.788	0.718
EfficientNet-b1	$7$M	0.832	0.850	0.832	0.716	0.803	0.716
DenseNet-169	$96$M	0.850	0.865	0.850	0.718	0.794	0.718
ResNet-152	$84$M	0.829	0.866	0.829	0.705	0.784	0.705
Inception-v3	$74$M	0.871	0.884	0.871	0.737	0.782	0.737
DarkNeta[34]	$1.94$M	0.712	0.699	0.712	0.495	0.245	0.495
CoroNeta[35]	$124$M	0.869	0.877	0.869	0.689	0.776	0.689
Proposed CVR-Net	$48$M	$\mathbf{0.887}$	$\mathbf{0.885}$	$\mathbf{0.887}$	$\mathbf{0.799}$	$\mathbf{0.821}$	$\mathbf{0.799}$

^aWe have implemented those models in our experimental settings for ablation studies.