Table 6.

Detailed comparison of the proposed method in the cross-dataset setting. The proposed method includes two encoder-decoder networks, referred to as Network-1 and Network-2, respectively.

Methods	Train dataset	Test dataset	Dice	IoU
DED-CNN	JSRT	SCXR	0.9525	0.9108
DED-CNN w/o SE	JSRT	SCXR	0.9381	0.8855
DED-CNN w/o AGM	JSRT	SCXR	0.9399	0.8869
Network-1	JSRT	SCXR	0.9416	0.8913
Network-2	JSRT	SCXR	0.9480	0.9014
DED-CNN	SCXR	MCCXR	0.9597	0.9228
DED-CNN w/o SE	SCXR	MCCXR	0.9478	0.9013
DED-CNN w/o AGM	SCXR	MCCXR	0.9550	0.9142
Network-1	SCXR	MCCXR	0.9495	0.9068
Network-2	SCXR	MCCXR	0.9559	0.9167
DED-CNN	SCXR+MCCXR	NIH	0.9168	0.8470
DED-CNN w/o AGM	SCXR+MCCXR	NIH	0.9148	0.8436
Network-1	SCXR+MCCXR	NIH	0.9051	0.8383
Network-2	SCXR+MCCXR	NIH	0.9081	0.8433

The bolded values represent the best performances.