Table 2.
Class-wise precision performance comparison with other deep learning techniques in literature with our findings for COVID-19 detection
| Backbone | Accuracy | COVID-19 | Normal | Pneumonia |
|---|---|---|---|---|
| Concurrent proposed approach: | ||||
| VGG16 [16] | 0.77 | 0.636 | – | – |
| COVIDNet-CXR Small [23] | – | 0.964 | 0.898 | 0.947 |
| Flat - EfficientNetB0 [16] | 0.90 | 1.0 | – | – |
| Flat - EfficientNetB3 [16] | 0.939 | 1.0 | – | – |
| COVIDNet-CXR Large [23] | 0.943 | 0.909 | 0.917 | 0.989 |
| COVIDNet-CXR3-A[23] | – | 0.979 | 0.921 | 0.903 |
| ResNet18 [2] | 0.951 | 0.918 | 0.943 | – |
| Our results: | ||||
| VGG16 (v1 Augmentation) | 0.88 | 0.82 | 0.84 | 0.98 |
| VGG16 (v2 GAN Augmentation) | 0.90 | 0.93 | 0.87 | 0.96 |
| Resnet50 (v2 Augmentation) | 0.943 | 0.97 | 0.93 | 0.96 |
| EfficientNetB0 (v2 Augmented) | 0.968 | 1.0 | 0.96 | 0.96 |