Table 6.

Performance metrics achieved using combinations of the augmented training data toward classifying Twitter and Montreal COVID-19 CXR collections as belonging to the viral pneumonia category. Bold values indicate superior performance.

Dataset	Accuracy
	Twitter-COVID-19	Montreal-COVID-19
Baseline	0.2885	0.5028
Data augmentation with weakly labeled images
Baseline + NIH	0.1037	0.2625
Baseline + CheXpert	0.5555	0.6536
Baseline + RSNA	0.2296	0.4469
Baseline + NIH + CheXpert	0.1852	0.4078
Baseline + NIH + RSNA	0.1407	0.4413
Baseline + CheXpert + RSNA	0.2222	0.4357
Baseline + NIH + CheXpert + RSNA	0.1852	0.4413
Baseline + CheXpert + Twitter	-	0.7095
Baseline + CheXpert + Montreal	0.8889	-
Baseline + Twitter	-	0.9778
Baseline + Montreal	0.9926	-