Table 4.

Comparison of AUC against application of CheXNet⁴⁷ with the PadChest dataset

	Atelectasis	Cardiomegaly	Consolidation	Oedema	Lesion	Pneumonia	Pneumothorax	No finding
CheXNet—100%	0.794	0.908	0.840	0.939	0.707	0.806	0.873	0.871
CheXzero—0%	0.798	0.898	0.871	0.961	0.651	0.789	0.766	0.755

The pathologies selected are the pathologies reported in CheXNet⁴⁷. The self-supervised method that is trained on a dataset from a country different from PadChest outperforms CheXNet, a fully supervised model trained on PadChest, on three out of eight pathologies reported in CheXNet: atelectasis, consolidation and oedema. Numbers within parentheses indicate 95% CI.