Table 3.
Performance achieved by the custom WRN and pretrained CNNs in classifying the pediatric CXR dataset into bacterial and viral categories. Here Acc.: Accuracy, Sens.: Sensitivity, Prec.: Precision, F: F-score, and MCC: Matthews Correlation Coefficient). Here bold values indicate superior performance.
| Models | Acc. | AUC | Sens. | Spec. | Prec. | F | MCC |
|---|---|---|---|---|---|---|---|
| Custom WRN | 0.8974 | 0.9534 | 0.9381 | 0.8311 | 0.9008 | 0.9191 | 0.7806 |
| VGG-16 | 0.9308 | 0.9565 | 0.9711 | 0.8649 | 0.9216 | 0.9457 | 0.8527 |
| Inception-V3 | 0.9103 | 0.937 | 0.9587 | 0.8311 | 0.9028 | 0.9299 | 0.8085 |
| Xception | 0.9282 | 0.954 | 0.9546 | 0.8852 | 0.9315 | 0.9429 | 0.8469 |
| DenseNet-121 | 0.9026 | 0.9408 | 0.967 | 0.7973 | 0.8864 | 0.925 | 0.7931 |
| NASNet-mobile | 0.9282 | 0.9479 | 0.9753 | 0.8514 | 0.9148 | 0.944 | 0.8477 |