Table 4.
Comparison of algorithms of the existing methods for the automated diagnosis of COVID19.
| Ref. no. | Method | Classification scheme | Accuracy | Recall/sensitivity | CV |
|---|---|---|---|---|---|
| Proposed model | COVID-19 vs. Pneumonia vs. No-findings (Dataset A) |
96% | 96.67% | 5-Fold | |
| COVID-19 vs. No-findings (Dataset A) |
100% | 100% | 5-Fold | ||
| COVID-19 vs. Pneumonia vs. No-findings (Dataset B) |
97.17% | 97.17% | 5-Fold | ||
| COVID-19 vs. No-findings (Dataset B) |
96.06% | 96% | 5-Fold | ||
| [18] | COVID-CAPS | COVID vs. non-COVID | 95.7% | 90% | Hold out |
| [19] | InceptionV3 | COVID-19 vs. No-finding | 96.2% | 97.1% | 5-Fold |
| ResNet50 | COVID-19 vs. No-finding | 96.1% | 91.8% | 5-Fold | |
| ResNet101 | COVID-19 vs. No-finding | 96.1% | 78.3% | 5-Fold | |
| ResNet152 | COVID-19 vs. No-finding | 93.9% | 65.4% | 5-Fold | |
| Inception-ResNetV2 | COVID-19 vs. No-finding | 94.2% | 83.5% | 5-Fold | |
| [20] | ResNet50 plus SVM | COVID-19 vs. Pneumonia vs. No-finding | 95.33% | 95.33% | Hold out |
| [21] | VGG-19 | COVID-19 vs. Pneumonia vs. No-finding | 82.24% | 83% | Hold out |
| ResNet-50 | COVID-19 vs. Pneumonia vs. No-finding | 90.67% | 90.6% | Hold out | |
| COVID-Net | COVID-19 vs. Pneumonia vs. No-finding | 93.34% | 93.3% | Hold out | |
| [22] | COVIDX-Net | COVID-19 vs. No-finding | 90% | 90% | Hold out |
| [23] | Transfer learning with convolutional neural networks | COVID-19 vs. Pneumonia vs. No-finding | 94.72% | 98.66% | 10-Fold |
| [25] | DarkCovidNet | COVID-19 vs. Pneumonia vs. No-finding | 87.02% | 85.35% | 5-Fold |
| COVID-19 vs. No-finding | 98.08% | 95.13% | 5-Fold | ||
| [26] | EfficientNet-B0 | COVID-19 vs. Pneumonia vs. No-finding | 95.24% | 93.61% | Hold out |
| 2D curvelet transform-EfficientNet-B0 | COVID-19 vs. Pneumonia vs. No-finding | 96.87% | 95.68% | Hold out | |
| [27] | Inception-V3 | COVID-19 vs. Pneumonia vs. No-finding | 85% | 94% | 5-Fold |