Table 6.
Performance of the FCONet framework based on institutional data split for COVID-19 data.
| Model and data group | Sensitivity, % | Specificity, % | Accuracy, % | AUCa | P value | ||
| ResNet-50 | <.001 | ||||||
|
|
COVID-19 pneumonia | 97.39 | 99.64 | 98.67 | 0.99 |
|
|
|
|
Other pneumonia | 99.26 | 98.45 | 98.637 | 0.99 |
|
|
|
|
Nonpneumonia | 100 | 100 | 100 | 1.0 |
|
|
| VGG16 | <.001 | ||||||
|
|
COVID-19 pneumonia | 97.15 | 99.64 | 98.57 | 0.99 |
|
|
|
|
Other pneumonia | 99.26 | 98.31 | 98.57 | 0.99 |
|
|
|
|
Nonpneumonia | 100 | 100 | 100 | 1.0 |
|
|
| Xception | <.001 | ||||||
|
|
COVID-19 pneumonia | 90.50 | 94.82 | 92.97 | 0.98 |
|
|
|
|
Other pneumonia | 89.30 | 94.37 | 92.97 | 0.98 |
|
|
|
|
Nonpneumonia | 100 | 100 | 100 | 1.0 |
|
|
| Inception-v3 | <.001 | ||||||
|
|
COVID-19 pneumonia | 74.58 | 99.46 | 88.79 | 0.98 |
|
|
|
|
Other pneumonia | 97.42 | 84.93 | 88.38 | 0.97 |
|
|
|
|
Nonpneumonia | 100 | 99.42 | 99.59 | 0.99 |
|
|
aAUC: area under the curve.