Table 2 Performance comparison of K-MIL against various baselines on the breast cancer dataset comprising of H&E stained images

5-Random net

0.684 ± 0.19

0.690 ± 0.32

0.540 ± 0.470

0.658 ± 0.38

0.670 ± 0.23

gated ABMIL

0.745 ± 0.11

0.795 ± 0.20

0.673 ± 0.2

0.728 ± 0.20

0.845 ± 0.11

ABMIL

0.762 ± 0.10

0.777 ± 0.21

0.725 ± 0.21

0.75 ± 0.19

0.844 ± 0.11

Mi-NET

0.707 ± 0.64

0.707 ± 0.18

0.619 ± 0.27

0.839 ± 0.02

0.712 ± 0.003

MI-NET

0.724 ± 0.10

0.730 ± 0.10

0.763 ± 0.17

0.746 ± 0.12

0.888 ± 0.09

MI-NET with RC

0.755 ± 0.28

\(0.738\pm\)0.11

0.725 ± 0.21

0.731 ± 0.14

0.855 ± 0.12

MI-NET with DS

0.734 ± 0.12

0.736 ± 0.18

0.716 ± 0.18

0.728 ± 0.18

0.847 ± 0.10

Ours (euclidean)

0.890 ± 0.07

0.943 ± 0.08

0.821 ± 0.16

0.877 ± 0.11

0.970 ± 0.07

Ours (siamese)

0.910 ± 0.08

0.931 ± 0.12

0.869 ± 0.16

0.898\(\pm\)0.14

0.977 ± 0.13