Table 2 Metric values (95% CI) at the selected operating point for H&E BE-TransMIL and TFF3 BE-TransMIL models on the discovery dataset

From: Enabling large-scale screening of Barrett’s esophagus using weakly supervised deep learning in histopathology

Metric

H&E BE-TransMIL

TFF3 BE-TransMIL

AUROC

0.914 (0.870–0.952)

0.939 (0.903–0.969)

AUPR

0.901 (0.813–0.925)

0.930 (0.850–0.946)

Accuracy

0.847 (0.803–0.900)

0.899 (0.855–0.934)

Sensitivity

0.727 (0.647–0.833)

0.791 (0.707–0.882)

Specificity

0.922 (0.874–0.963)

0.965 (0.919–0.986)

Back to article page