Table 3.

Comparison of performance on TEST using different data based on the configuration selection on VAL_PTEST. AUC for each class versus rest and WA-AUC are provided

	Only manually labeled data (TRAIN_PRI)	Manually and pseudo- labeled data (TRAIN_PRI + pseudo-labeled TRAIN_SEC)	p value
AUC normala	0.842	0.894	< 0.002
AUC abnormala	0.848	0.896	< 0.001
AUC arthroplasty	0.987	0.990	0.4006
WA-AUCa	0.857	0.903	0.001

^aSignificantly improved when training with manually and pseudo-labeled data compared to manually labeled data only