Table 5 Classification results for multiple ResNet-50 models, each fine-tuned, on points of interest (e.g. blood).

From: Galar - a large multi-label video capsule endoscopy dataset

Label

F1

AUROC

Accuracy

blood

0.14

0.87

0.98

pylorus

0.01

0.74

0.99

z-line

0.02

0.87

1.00

ulcer

0.00

0.44

0.99

polyp

0.05

0.73

1.00

erythema

0.02

0.70

1.00

  1. The metrics are computed individually for each label. The outcomes are averaged across the 2 cross-validation folds.
Back to article page