Table 3.

Dice score of each class calculated on the testsets of the binary subsets, $\mu$ being the mean over all classes

Trial	Abd	Col	Liv	Pan	Sma	Sto	$\mu$
Separate models of EN	0.894	0.790	0.786	0.400	0.866	0.701	0.740
IL (ours)	0.892	0.782	0.741	0.364	0.849	0.709	0.723
IL w/o additional neg. labels	0.897	0.766	0.783	0.326	0.832	0.633	0.706

The first row shows the performance of each separate model in the ensemble before applying the argmax. The last row shows a variant of the IL approach if no additional negative samples are inferred. The classes Abdominal Wall, Colon, Liver, Pancreas, Small Intestine, and Stomach are abbreviated by their first three letters