Table 1.
Proposed architecture.
| Layer | Type | Parameters |
|---|---|---|
| Input | Image | 224 × 224 × 3 |
| Conv1 | Conv2D | 32 filters |
| Pool1 | MaxPool | 2 × 2 |
| Conv2 | Conv2D | 64 filters |
| Pool2 | MaxPool | 2 × 2 |
| Dense | FullyConnected | 256 |
| Output | Softmax | 4 classes |
Table 2.
Distribution of the dataset.
| No. of images | Class |
|---|---|
| 1281 | Covid-19 CXRs |
| 3270 | Normal CXRs |
| 1656 | Viral pneumonia CXRs |
| 3001 | Bacterial pneumonia CXRs |

Fig. 1.
Sample images from dataset.
Table 3.
Hyperparameter values against various optimization algorithms.
| Model | Dropout_1 | Filter count | Dropout_2 | Dense_activation | Dropout_3 | Learning rate | Units | Epochs | Score | Loss | Accuracy |
|---|---|---|---|---|---|---|---|---|---|---|---|
| GSO | 0.2 | 32 | 0.3 | softmax | 0.1 | 0.0059 | 32 | 25 | 0.1006 | 0.8929 | 0.7357 |
| RSO | 0.35 | 64 | 0.45 | softmax | 0.1 | 0.0052 | 192 | 45 | 0.1408 | 0.7412 | 0.7659 |
| BO | 0.05 | 256 | 0 | softmax | 0.15 | 0.002419 | 448 | 15 | 0.1789 | 0.4893 | 0.7492 |
| PSO | 0.25 | 128 | 0.25 | softmax | 0.07 | 0.00171507 | 260 | 40 | 0.2121 | 0.4893 | 0.8931 |
| EHO | 0.2 | 128 | 0.1 | softmax | 0.05 | 0.0038287 | 256 | 35 | 0.6556 | 0.296 | 0.9237 |
| Proposed AEHO-PRLS | 0.1 | 64 | 0.05 | softmax | 0.01 | 0.00024914 | 256 | 28 | 0.727 | 0.1586 | 0.9657 |

Fig. 2.
Performance comparison of the proposed model.
Table 4.
Ablation study of proposed model.
| Model | Accuracy [%] |
|---|---|
| CNN | 88.12 |
| CNN + EHO | 92.37 |
| CNN + PRLS | 93.12 |
| CNN + AEHO-PRLS | 96.57 |