Skip to main content

Table 4 Comparison of prediction performance from internal validation (healthcare center dataset) to detect epiretinal membrane in fundus photographs

From: Development of a generative deep learning model to improve epiretinal membrane detection in fundus photography

Classification network architecture

Augmentation

ROC-AUC (95% CI)

Sensitivity (%, 95% CI)

Specificity (%, 95% CI)

P-value

EfficientNetB0

StyleGAN2 (Ours)

0.926 (0.890–0.963)

92.0 (82.4–97.3)

80.8 (75.3–85.4)

Reference

EfficientNetB0

DDPM [26, 27]

0.825 (0.779–0.866)

88.9 (78.4–95.4)

65.2 (58.9–71.1)

0.0048

EfficientNetB0

CutMixa [28]

0.837 (0.792–0.877)

77.8 (65.5–87.3)

86.0 (81.1–90.1)

0.0080

Vision Transformer [29]

Classic augmentationa

0.835 (0.789–0.874)

84.1 (72.7–92.1)

75.2 (69.4–80.4)

0.0051

Vision Transformer [29]

StyleGAN2

0.863 (0.819–0.899)

82.5 (70.9–90.9)

84.0 (78.9–88.3)

0.0914

  1. CI confidence interval, DDPM denoising diffusion probabilistic model, ROC-AUC area under the receiver operating characteristic curve
  2. aWe oversampled the ERM class to balance the training dataset
Back to article page

BMC Medical Informatics and Decision Making

ISSN: 1472-6947

Contact us