Table 4.

COVID-19 recognition results from different experiments of multi-class classification (see in Table 1) applying the proposed network on CXR images employing 5-fold cross-validation.

Different studiesa	Dataset distribution	Metrics
	(Train$/$Val$/$Test)	Recall	Precision	Accuracy
	NOR: $951/316/316$	$0.925\pm 0.011$	$0.940\pm 0.009$	$0.925\pm 0.012$
	NCP: $2,565/854/854$	$0.978\pm 0.003$	$0.969\pm 0.006$	$0.977\pm 0.003$
	CVP: $300/100/100$	$0.944\pm 0.041$	$0.976\pm 0.010$	$0.946\pm 0.041$
CXR-Single-CL3	Weighted Average	$0.964\pm 0.005$	$0.963\pm 0.004$	$0.964\pm 0.005$
	NOR: $2,155/718/718$	$0.970\pm 0.018$	$0.844\pm 0.029$	$0.970\pm 0.018$
	NCP: $2,757/919/919$	$0.863\pm 0.029$	$0.990\pm 0.004$	$0.863\pm 0.029$
	CVP: $2,409/803/803$	$0.980\pm 0.008$	$0.968\pm 0.019$	$0.980\pm 0.008$
CXR-Multiple-CL3	Weighted Average	$0.933\pm 0.013$	$0.940\pm 0.011$	$0.933\pm 0.013$
	NOR: $2,155/718/718$	$0.962\pm 0.023$	$0.902\pm 0.026$	$0.962\pm 0.023$
	OBP: $1,668/556/556$	$0.741\pm 0.021$	$0.874\pm 0.023$	$0.741\pm 0.021$
	OVP: $897/298/298$	$0.705\pm 0.050$	$0.646\pm 0.032$	$0.705\pm 0.051$
	CVP: $2,409/803/803$	$0.975\pm 0.007$	$0.968\pm 0.011$	$0.975\pm 0.007$
CXR-Multiple-CL4	Weighted Average	$0.882\pm 0.003$	$0.886\pm 0.004$	$0.882\pm 0.003$

^aX-Y-CL#: X is CXR or CT; Y denotes the way images from different sources are combined for each class during training or evaluation; CL# is the number of classes. Details in Table 1.