Table 2.
Classification metrics of HAM10000 images at two different resolutions and without aggregated clinical features using eight different DNNs.
| Low-Resolution Images 1 | High-Resolution Images | |||
|---|---|---|---|---|
| DNN 2 | Accuracy 3 | Error Rate | Accuracy | Error Rate |
| ResNet34 | 75.32 | 24.68 | 76.73 | 23.27 |
| ResNet50 | 74.56 | 25.44 | 75.97 | 24.03 |
| ResNet101 | 75.62 | 24.38 | 77.02 | 22.98 |
| SEResNet50 | 77.82 | 22.22 | 79.13 | 20.87 |
| VGG16 | 76.85 | 23.15 | 78.25 | 21.75 |
| VGG19 | 74.21 | 25.79 | 75.62 | 24.38 |
| EfficientNetB5 | 74.05 | 25.91 | 75.50 | 24.5 |
| MobileNet | 82.47 | 17.53 | 83.88 | 16.12 |
1 low-resolution image: 300 × 224 RGB; high-resolution image: 300 × 224 RGB. 2 Deep neural network. 3 accuracy and error rates are expressed as percentages.