Table 3 Performance of the DMIL models with different backbone networks.

Backbone	Pooling	ACC	AUC	SEN	SPE	F1-Score
AlexNet	Max	0.848	0.937	0.862	0.867	0.847
	Conv	0.817	0.922	0.828	0.833	0.814
	AM	0.866	0.951	0.892	0.900	0.862
VGG16	Max	0.850	0.958	0.800	0.767	0.862
	Conv	0.817	0.931	0.788	0.767	0.825
	AM	0.883	0.982	0.871	0.867	0.885
ResNet34	Max	0.783	0.856	0.793	0.800	0.779
	Conv	0.800	0.886	0.781	0.767	0.806
	AM	0.816	0.864	0.771	0.733	0.831
DenseNet	Max	0.817	0.883	0.828	0.833	0.814
	Conv	0.783	0.892	0.775	0.767	0.787
	AM	0.850	0.927	0.889	0.900	0.842
MobileNetV2	Max	0.800	0.871	0.765	0.733	0.813
	Conv	0.817	0.907	0.771	0.733	0.831
	AM	0.850	0.963	0.784	0.733	0.866

*AM attention mechanism, ACC accuracy, AUC area under the curve, SEN sensitivity, SPE specificity.

Quick links

Search