Table 2.
Performances of different 3D networks on the PRM dataset, the same 3D-CNN on different input datasets, and different 2D networks on the PRM dataset.
| Accuracy | Precision | Sensitivity | F1 | Specificity | AUC score | |
|---|---|---|---|---|---|---|
| Different 3D networks with PRM dataset | ||||||
| 3D-CNN—Naive model | 89.3 | 82.6 | 88.3 | 85.1 | 93.6 | 0.937 |
| 3D-CNN—DenseNet121 | 86.9 | 78.1 | 85.8 | 81.2 | 92.3 | 0.904 |
| 3D-CNN—VGG16 | 77.4 | 69.8 | 58.5 | 57.4 | 84.0 | 0.827 |
| 3D-CNN—Resnet50 | 87.2 | 79.8 | 85.8 | 82.4 | 92.2 | 0.906 |
| 3D-CNN—InceptionV3 | 83.9 | 76.0 | 72.7 | 72.8 | 88.5 | 0.861 |
| 3D-CNN with different input datasets | ||||||
| IN | 85.1 | 80.6 | 71.8 | 74.8 | 88.2 | 0.900 |
| EX | 86.5 | 75.5 | 86.9 | 80.3 | 93.7 | 0.907 |
| ΔVairf | 86.6 | 80.0 | 80.4 | 79.9 | 90.1 | 0.897 |
| J | 85.2 | 76.4 | 83.3 | 79.4 | 91.1 | 0.895 |
| ADI | 83.0 | 74.1 | 78.0 | 75.2 | 88.8 | 0.862 |
| SRI | 84.9 | 76.8 | 80.4 | 77.7 | 90.2 | 0.886 |
| Concatenate of IN and EX CT images | 87.2 | 76.1 | 95.1 | 84.0 | 97.2 | 0.923 |
| Different 2D networks with PRM dataset | ||||||
| 2D-CNN—Naive model | 84.8 | 78.6 | 71.0 | 74.6 | 87.3 | 0.861 |
| Pretrained—DenseNet121 | 86.7 | 77.8 | 85.4 | 81.4 | 92.0 | 0.899 |
| Pretrained—VGG16 | 87.5 | 74.1 | 97.6 | 84.2 | 98.5 | 0.938 |
| Pretrained—Resnet50 | 88.3 | 84.6 | 80.5 | 82.5 | 90.1 | 0.901 |
| Pretrained—InceptionV3 | 88.3 | 80.0 | 87.8 | 83.7 | 93.3 | 0.923 |