Table 3 Performance of models in the prospective validation dataset
Model number | Parameters | Architecture | Accuracy (95% CI) | Sensitivity (95% CI) | Specificity (95% CI) | AUC (95% CI) |
|---|---|---|---|---|---|---|
1 | Class weight = [0.1, 25] Epoch = 80 | ResNet101 | 0.718 (0.556–0.838) | 0.800 (0.548–0.930) | 0.667 (0.454–0.828) | 0.880 (0.736–0.951) |
2 | ResNet50 | 0.833 (0.681–0.921) | 0.733 (0.481–0.891) | 0.905 (0.711–0.974) | 0.930 (0.799–0.978) | |
3 | InceptionResNetV2 | 0.806 (0.650–0.903) | 0.933 (0.702–0.988) | 0.714 (0.500-0.862) | 0.867 (0.720–0.943) | |
4 | InceptionV3 | 0.778 (0.619–0.883) | 0.800 (0.548–0.930) | 0.762 (0.549–0.894) | 0.903 (0.765–0.964) | |
5 | Class weight = [0.1, 30] Epoch = 60 | ResNet101 | 0.889 (0.747–0.956) | 0.933 (0.702–0.988) | 0.857 (0.654–0.950) | 0.966 (0.850–0.993) |
6 | ResNet50 | 0.750 (0.589–0.863) | 0.867 (0.621–0.963) | 0.667 (0.454–0.828) | 0.872 (0.726–0.946) | |
7 | InceptionResNetV2 | 0.833 (0.681–0.921) | 0.933 (0.702–0.988) | 0.762 (0.549–0.894) | 0.954 (0.832–0.989) | |
8 | InceptionV3 | 0.778 (0.619–0.883) | 1.000 (0.796-1.000) | 0.619 (0.409–0.793) | 0.871 (0.726–0.946) |