Table 2.
Results comparisons of the proposed COVID-WideNet to other state-of-the-art models.
| Architecture | Accuracy | Sensitivity | Specificity | AUC | Parameters (in Millions) |
Pre-trained on | Data augmentation |
|---|---|---|---|---|---|---|---|
| COVID-CAPS [27] | 98.3% | 80% | 98.6% | 0.99 | 0.29M | NIH Chest X-ray dataset [56] | False |
| VGG-19 [36] | 83.0% | 58.7% | – | – | 20.37M | ImageNet [57] | True |
| ResNet-50 [36] | 90.6% | 83.0% | – | – | 24.97M | ImageNet [57] | True |
| COVID-Net [36] | 93.3% | 91% | – | – | 11.75M | ImageNet [57] | True |
| VGG-CapsNet [30] | 97% | – | – | 0.96 | – | X rays and CTS | – |
| DenseCapsNet [31] | 90.7% | – | 95.3% | 0.93 | – | CXR images | – |
| CT-Caps [32] | 90.8% | 94.5% | 86% | – | – | COVID-CT-MD | – |
| COVID-WideNet | 91% | 91% | 91.14% | 0.95 | 0.43M | No pre-training | False |