Table 3 Comparison and correlation of the ConvNetUHCMC/CWRU and ConvNetHUP classifiers in terms of Dice, PPV, NPV, TPR, TNR, FPR and FNR.

From: Accurate and reproducible invasive breast cancer detection in whole-slide images: A Deep Learning approach for quantifying tumor extent

 

Dice

PPV

NPV

TPR

TNR

FPR

FNR

TCGA

 ConvNetHUP

0.7494 ± 0.2071

0.7071 ± 0.2254

0.9658 ± 0.0514

0.8600 ± 0.1705

0.9188 ± 0.0805

0.0812 ± 0.0805

0.1400 ± 0.1705

 ConvNetUHCMC/CWRU

0.7068 ± 0.2061

0.6464 ± 0.2188

0.9629 ± 0.0584

0.8676 ± 0.1706

0.8880 ± 0.0824

0.1120 ± 0.0824

0.1324 ± 0.1706

 r

0.8733

0.9258

0.8109

0.6345

0.8055

0.8055

0.6345

NC

 ConvNetHUP

N/A

N/A

1 ± 0

N/A

0.9716 ± 0.0693

0.0284 ± 0.0693

N/A

 ConvNetUHCMC/CWRU

N/A

N/A

1 ± 0

N/A

0.9546 ± 0.0816

0.0454 ± 0.0816

N/A

 r

N/A

N/A

N/A

N/A

0.6876

0.6876

N/A

  1. Note that for the normal cases considered, not all the performance measures are shown because the NC data cohort did not have cancer annotations.
Back to article page