Table 1 Performance evaluation of internal test set for CheXNet model.

From: Automation of generative adversarial network-based synthetic data-augmentation for maximizing the diagnostic performance with paranasal imaging

Performance metrics for model predictions

ODa

GCBb

GCB + CDAc

GCB + OMGDAd

Accuracy

0.785

0.798

0.81

0.899

Sensitivity

0.767

0.861

0.791

0.907

Specificity

0.806

0.722

0.833

0.889

F1-score

0.795

0.822

0.819

0.907

PPV

0.825

0.787

0.850

0.907

NPV

0.744

0.813

0.769

0.889

AUC

0.855

0.887

0.908

0.963

  1. aCheXNet model trained using original data, only.
  2. bCheXNet model trained with GAN-based class balancing.
  3. cCheXNet model trained with GAN-based class balancing and conventional data augmentation.
  4. dCheXNet model trained with the output of the suggested automation pipeline, which is the trained model of Koptimal (GCB + OMGDA).
Back to article page