Table 5.
Evaluation of COVID-19 detection system with different state-of-the-art methods.
| Number of subjects | Method | Sensitivity (%) | Specificity (%) | Accuracy (%) | F1-score (%) |
|---|---|---|---|---|---|
| 349 COVID-19397 non-COVID-19 | Self-supervised learning with transfer learning, DenseNet-201 [10] | NA | NA | 86 | 85 |
| 349 COVID-19 463 non-COVID-19 |
Multi-tasking learning approach [11] | NA | NA | 89 | 90 |
| 349 COVID-19 397 non-COVID-19 |
Different CNN models AlexNet, VGGNet16, VGGNet19, GoogleNet, ResNet50 [12] | NA | NA | 82.91 | NA |
| 564 COVID-19 660 non-COVID-19 |
VGG16 based lesion attention DNN [13] | 88.80 | NA | 88.60 | 87.9 |
| 313 COVID-19 229 non-COVID-19 |
UNet [15] | 90.70 | 91.1 | 90.10 | NA |
| 413 COVID-19 439 non-COVID-19 |
ResNet-50 + 2D CNN [16] | 91.46 | 94.78 | 93.02 | NA |
| 230 COVID-19 130 normal |
AD3D-MIL [17] | 97.90 | NA | 97.90 | 97.90 |
| 1029 COVID-19 1695 non-COVID-19 |
AH-Net DenseNet-201 [18] | 84.0 | 93.0 | 90.80 | NA |
| 496 COVID-19 1385 others |
CNN [21] | 94.06 | 95.47 | 94.98 | NA |
| 349 COVID-19 397 non-COVID-19 |
ResNet18 [22] | 100 | 98.60 | 99.40 | 99.5 |
| 760 COVID-19 by augmentation 736 non-COVID-19 |
ResNet-50 with data augmentation [proposed work] | 98.58 | 98.40 | 98.5 | 98.58 |