Table 3.

Test performance of the aggregation network when using different information combinations as inputs. Models utilizing both local and global information achieved better performance than the counterparts using single type of maps

	AUC
location indicator maps	0.474 ± 0.031
embedding maps	0.778 ± 0.002
saliency maps	0.695 ± 0.003
location indicator maps + embedding maps	0.777 ± 0.002
location indicator maps + saliency maps	0.721 ± 0.011
embedding maps + saliency maps	0.799 ± 0.002
location indicator maps + saliency maps + embedding maps	0.797 ± 0.001