Table 1.

Evaluation of the proposed methods on YUMC dataset including results of using DA-CMIL with/without contrastive loss ${\mathcal{L}}_F$.

Method	Accuracy	AUC	F1	Specificity	Sensitivity
DeCoVNet Wang et al. (2020c)	0.831	0.825	0.8	0.875	0.774
MIL Campanella et al. (2019)	0.803	0.796	0.767	0.85	0.742
DeepAttentionMIL Ilse et al. (2018)	0.859	0.875	0.861	0.75	1
JointMIL Chikontwe et al. (2020)	0.901	0.909	0.896	0.85	0.968
Zhang3DCNN Zhang et al. (2020b)	0.93	0.938	0.925	0.875	1
DA-CMIL (w/o ${\mathcal{A}}_{\theta ,S,I}$)	0.76	0.72	0.62	1.0	0.45
DA-CMIL (w/o ${\mathcal{L}}_F$)	0.93	0.934	0.923	0.9	0.968
DA-CMIL (w/ ${\mathcal{L}}_F$)	0.986	0.984	0.984	0.975	1