Skip to main content

Table 1 Performance comparison of K-MIL against various baselines on the COLON cancer dataset comprising of H&E stained images.

From: Not seeing the trees for the forest. The impact of neighbours on graph-based configurations in histopathology

METHOD

ACCURACY

PRECISION

RECALL

F-SCORE

AUC

5-Random net

0.781 ± 0.11

0.774 ± 0.16

0.799± 0.13

0.786±0.14

0.76±0.07

gated ABMIL net

0.905 ± 0.08

0.892 ± 0.15

0.911±0.10

0.898±0.18

0.985±0.03

ABMIL net

0.911 ± 0.08

0.921 ± 0.12

0.905±0.15

0.912±0.13

0.987±0.02

MI-NET

0.809 ± 0.129

0.841 ± 0.182

0.813±0.21

0.925± 0.02

0.925 ± 0.09

Mi-NET

0.842 ± 0.02

0.866 ± 0.01

0.816 ± 0.03

0.839 ± 0.02

0.914 ± 0.01

MI-NET with RC

0.879 ± 0.11

\(0.820\pm\)0.16

0.950±0.15

0.880±0.15

0.975±0.004

MI-NET with DS

0.853 ± 0.13

0.794 ± 0.27

0.853±0.28

0.822±0.27

0.959±0.07

Ours (euclidean)

0.909 ± 0.10

0.923 ± 0.12

0.925± 0.12

0.920±0.12

0.974±0.05

Ours (siamese)

0.934±0.08

0.946±0.09

0.930±0.13

0.937±0.09

0.987±0.07

  1. The experiments were run 5 times and the average (± standard error of the mean) is reported. [bold]: Highlights the best-performing results in the respective metrics