Table 5.
Summary of model performance for classification of non-Covid pulmonary infections. The best performing model in each metric is highlighted in green. EfficientNetB5 attained the highest F1 score and accuracy, DenseNet201 the greatest sensitivity, and EfficientNetB6 the highest specificity and precision.
| # | Model | F1 | Accuracy | Sensitivity | Specificity | Precision |
|---|---|---|---|---|---|---|
| 1 | DenseNet121 | 0.8239 ± 0.0242 | 0.8966 ± 0.0131 | 0.8188 ± 0.0313 | 0.9293 ± 0.0100 | 0.8315 ± 0.0234 |
| 2 | DenseNet169 | 0.8245 ± 0.0244 | 0.8971 ± 0.0128 | 0.8188 ± 0.0313 | 0.9301 ± 0.0108 | 0.8333 ± 0.0243 |
| 3 | DenseNet201 | 0.8262 ± 0.0256 | 0.8989 ± 0.0133 | 0.8178 ± 0.0344 | 0.9325 ± 0.0091 | 0.8374 ± 0.0213 |
| 4 | EfficientNetB0 | 0.8207 ± 0.0249 | 0.8980 ± 0.0129 | 0.7976 ± 0.0300 | 0.9401 ± 0.0088 | 0.8483 ± 0.0233 |
| 5 | EfficientNetB1 | 0.7482 ± 0.0535 | 0.8623 ± 0.0244 | 0.7210 ± 0.0571 | 0.9221 ± 0.0180 | 0.7889 ± 0.0463 |
| 6 | EfficientNetB2 | 0.8121 ± 0.0246 | 0.8931 ± 0.0127 | 0.7901 ± 0.0311 | 0.9363 ± 0.0095 | 0.8398 ± 0.0233 |
| 7 | EfficientNetB3 | 0.8170 ± 0.0243 | 0.8952 ± 0.0129 | 0.7979 ± 0.0310 | 0.9360 ± 0.0099 | 0.8414 ± 0.0234 |
| 8 | EfficientNetB4 | 0.8288 ± 0.0247 | 0.9009 ± 0.0138 | 0.8143 ± 0.0301 | 0.9373 ± 0.0131 | 0.8496 ± 0.0280 |
| 9 | EfficientNetB5 | 0.8385 ± 0.0278 | 0.9077 ± 0.0140 | 0.8172 ± 0.0367 | 0.9458 ± 0.0084 | 0.8643 ± 0.0225 |
| 10 | EfficientNetB6 | 0.8157 ± 0.0200 | 0.8963 ± 0.0103 | 0.7747 ± 0.0273 | 0.9483 ± 0.0064 | 0.8648 ± 0.0166 |
| 11 | EfficientNetB7 | 0.8038 ± 0.0210 | 0.8856 ± 0.0111 | 0.7905 ± 0.0277 | 0.9262 ± 0.0106 | 0.8235 ± 0.0235 |
| 12 | InceptionResNetV2 | 0.7919 ± 0.0239 | 0.8790 ± 0.0123 | 0.7798 ± 0.0303 | 0.9210 ± 0.0099 | 0.8073 ± 0.0245 |
| 13 | InceptionV3 | 0.7963 ± 0.0286 | 0.8824 ± 0.0150 | 0.7799 ± 0.0367 | 0.9254 ± 0.0123 | 0.8177 ± 0.0275 |
| 14 | ResNet101V2 | 0.7837 ± 0.0279 | 0.8717 ± 0.0163 | 0.7818 ± 0.0311 | 0.9101 ± 0.0163 | 0.7900 ± 0.0340 |
| 15 | ResNet152V2 | 0.7835 ± 0.0254 | 0.8766 ± 0.0123 | 0.7600 ± 0.0381 | 0.9258 ± 0.0115 | 0.8154 ± 0.0244 |
| 16 | ResNet50 | 0.8177 ± 0.0241 | 0.8933 ± 0.0128 | 0.8114 ± 0.0323 | 0.9279 ± 0.0096 | 0.8272 ± 0.0232 |
| 17 | ResNet50V2 | 0.7697 ± 0.0275 | 0.8668 ± 0.0138 | 0.7552 ± 0.0355 | 0.9137 ± 0.0118 | 0.7888 ± 0.0273 |
| 18 | VGG16 | 0.6865 ± 0.0299 | 0.8240 ± 0.0136 | 0.6544 ± 0.0392 | 0.8959 ± 0.0146 | 0.7304 ± 0.0305 |
| 19 | VGG19 | 0.6346 ± 0.0303 | 0.7840 ± 0.0162 | 0.6011 ± 0.0692 | 0.8613 ± 0.0208 | 0.6535 ± 0.0284 |
| 20 | Xception | 0.8017 ± 0.0299 | 0.8854 ± 0.0159 | 0.7863 ± 0.0360 | 0.9268 ± 0.0122 | 0.8210 ± 0.0289 |