Table 7 Detection results on EOAD-I and ROMCI-I using different encoders
From: Early detection of dementia through retinal imaging and trustworthy AI
Dataset | Encoder | Accuracy | Precision | F1-score | Kappa | AUC |
|---|---|---|---|---|---|---|
ROAD-I | ResNet18 | 0.8885 | 0.8862 | 0.8867 | 0.7018 | 0.9355 |
ResNet50 | 0.8921 | 0.8902 | 0.8883 | 0.7027 | 0.9365 | |
DenseNet | 0.8914 | 0.8891 | 0.8881 | 0.7018 | 0.9471 | |
EfficientNet | 0.8790 | 0.8833 | 0.8806 | 0.6954 | 0.9330 | |
ConvNeXt | 0.8914 | 0.8891 | 0.8881 | 0.7018 | 0.9471 | |
ViT | 0.7744 | 0.7524 | 0.7514 | 0.3109 | 0.6902 | |
ROMCI-I | ResNet18 | 0.8487 | 0.8506 | 0.8410 | 0.6229 | 0.8630 |
ResNet50 | 0.8527 | 0.8702 | 0.8405 | 0.6254 | 0.8607 | |
DenseNet | 0.8320 | 0.8314 | 0.8237 | 0.5836 | 0.8785 | |
EfficientNet | 0.8208 | 0.8192 | 0.8119 | 0.5564 | 0.8124 | |
ConvNeXt | 0.8671 | 0.8747 | 0.8588 | 0.6608 | 0.8732 | |
ViT | 0.7946 | 0.7984 | 0.7964 | 0.3381 | 0.7399 |
