Table 4.
VF Prediction Accuracy of Five Models Using Original and Artifact-Corrected Datasets as Input
| Model | Original Dataset (Control) | Artifact-Corrected Dataset |
|---|---|---|
| DenseNet-121 | ||
| RMSE | 4.77 (4.40, 5.13) | 4.79 (4.42, 5.15) |
| MAE | 3.82 (3.50, 4.15) | 3.83 (3.50, 4.15) |
| ResNet-34 | ||
| RMSE | 4.65 (4.30, 4.99) | 4.71 (4.35, 5.07) |
| MAE | 3.69 (3.39, 3.99) | 3.72 (3.41, 4.03) |
| VGG-16 | ||
| RMSE | 4.97 (4.61, 5.33) | 5.03 (4.67, 5.39) |
| MAE | 3.94 (3.62, 4.26) | 3.97 (3.66, 4.29) |
| ViT-Base | ||
| RMSE | 4.57 (4.19, 4.96) | 4.50 (4.13, 4.88) |
| MAE | 3.55 (3.18, 3.82) | 3.50 (3.18, 3.82) |
| DINO-ViT Base | ||
| RMSE | 4.59 (4.22, 4.96) | 4.44 (4.07, 4.82) |
| MAE | 3.61 (3.29, 3.94) | 3.46 (3.14, 3.79) |
dB, decibels; DINO, Distillation with No Labels; MAE, mean absolute error; RMSE, root mean square error; VF, visual field; ViT, Vision Transformer.
Values are reported as mean RMSE (dB) or MAE (dB) along with 95% confidence intervals.