Table 2: Classification definitions between predictions and data set.

Each classification prediction made is evaluated against its “ground truth” label. Cases which agree are considered “concordant” and cases which disagree are considered “discordant.” This deviates from classical nomenclature of true/false positive/negative because of some uncertainty of pneumonia diagnosis in the training and test data set.

	Prediction	“Ground truth” label
Concordant Positives	Pneumonia	Pneumonia
Concordant Negatives	No pneumonia	No pneumonia
Discordant Positives	Pneumonia	No pneumonia
Discordant Negatives	No pneumonia	Pneumonia