Table 1.
Summary of automated COVID-19 detection systems developed. Unless stated otherwise, all accuracy results are reported according to 3 class classification (Normal, COVID-19, and Pneumonia) (2-Class and multi-class).
| Study | Dataset(s) | Classes | Classifier | Accuracy |
|---|---|---|---|---|
| Ozturk et al. 2020 | CO19-Ximage (Cohen, 2020) and Ch-X8image (Wang et al. 2017). |
Normal (5 0 0)COVID-19 (1 2 7) Pneumonia (5 0 0) |
Darknet-19 | 2-Class: 98.08% 3-Class: 87.02% |
| Li et al. 2021 | GitHub and Kaggle | Normal (2 3 4)COVID-19 (87) |
SVM | 100% |
| Asif et al. 2020 | CO19-Ximage (Cohen, 2020) and COVQU (Chowdhury et al., 2020, Rahman et al., 2021) | Normal (1,341)COVID-19 (8 6 4)Pneumonia (1,345) |
Inception-V3. | 98.30% |
| Brunese et al. 2020 | CO19-Ximage (Cohen, 2020), X-Ray Image Dataset (Ozturk et al. 2020), and Ch-X8image (Wang et al. 2017) | Normal (3,520) COVID-19 (2 5 0)VGG16.Pneumonia (2,753) |
VGG16. |
97.00% |
| Das et al. 2020 | X-Ray Image Dataset (Ozturk et al. 2020) | 1,000 chest X-rays images included Normal, COVID-19, and Pneumonia classes. | Xception model | 97.4 |
| Toraman et al. 2020 | CO19-Ximage (Cohen, 2020) and Ch-X8image (Wang et al. 2017) | Normal (1,050)COVID-19 (2 3 1)Pneumonia (1,050) |
capsule neural network | 2-Class: 97.24% 3-Class: 84.22% |
| Zhang et al. 2020 | CO19-Ximage (Cohen, 2020), Ch-X8image (Wang et al. 2017), X-Ray Image Dataset (Ozturk et al. 2020), and Kaggle. |
Normal (5 5 7)COVID-19 (2 3 4)Pneumonia (7 3 0) |
Inception-V3 | 90.00% |
| Abraham et al. 2020 | CO19-Ximage (Cohen, 2020) and Ch-Ximage (Mooney, 2020) | COVID-19 (4 5 3)non-COVID (4 9 7) |
Squeezenet + Darknet-53 + MobilenetV2 + Xception + Shufflenet |
2-Class: 91.16% |
| Jain et al. 2020 | CO19-Ximage (Cohen et al. 2020) and Ch-Ximage (Mooney, 2020) | Normal (3 1 5)COVID-19 (2 5 0)Bacterial Pneumonia (3 0 0)Viral Pneumonia (3 5 0) |
ResNet50 and ResNet-101 | Multi-class: 97.77% |
| Afshar et al. 2020 | CO19-Ximage (Cohen, 2020) and Ch-Ximage (Mooney et al. 2020) | 94,323 chest X-rays images included Normal, COVID-19, Bacterial Pneumonia, and Viral Pneumonia classes. | Capsule Networks | Multi-Class: 95.70% |
| Heidari et al. 2020 | Mendeley Data (Kermany et al. 2018), COVQU (Chowdhury et al., 2020, Rahman et al., 2021), and CO19-Ximage (Cohen, 2020) | Normal (2,880)COVID-19 (4 1 5)Pneumonia (5,179) |
VGG16 | 96.90% |
| Ismael and Şengür, 2021 | CO19-Ximage (Cohen, 2020) and CO19-Ximage (Mooney, 2020). | Normal (2 0 0)COVID-19 (1 8 0) |
ResNet50 + SVM classifier with the Linear kernel function | 2-Class: 94.70% |
| Jin et al. 2021 | COVQU (Chowdhury et al., 2020, Rahman et al., 2021) and CO19-Ximage (Cohen, 2020) | Normal (6 0 0)COVID-19 (5 4 3)Pneumonia (6 0 0) |
AlexNet + ReliefF + SVM | 99.43% |
| Demir, 2021 | Ch-Ximage (Mooney, 2020) and Ch-X8image (Wang et al. 2017) | Normal (2 0 0)COVID-19 (3 6 1)Pneumonia (5 0 0) |
LSTM | 97.11% |
| Sharifrazi et al. 2021 | Omid Hospital in Tehran | Normal (2 5 6)COVID-19 (77) |
CNN + SVM + Sobel filter |
2-Class: 99.02% |
| Quan et al. 2021 | CoronaHack (Praveen, 2020) CO19-Ximage (Cohen, 2020), and COVQU [7, 8] |
Normal (2,917)COVID-19 (7 8 1)Bacterial Pneumonia (2,850)Viral Pneumonia (2,884) |
DenseNet and CapsNet | 90.70% |
| Júnior et al. 2021 | CO19-Ximage (Cohen, 2020) and Ch-Ximage (Mooney 2020) | Normal (2 5 0)COVID-19 (2 5 0) |
CNN + PCA | 2-Class: 97.60–100% |
| Das et al. 2021 | Kaggle | Normal (1,341)COVID-19 (2 1 9)Pneumonia (1,345) |
VGG-16 and ResNet-50 | 97.67% |
| Albahli et al. 2021 | Ch-Ximage datasets (Ahsan et al. 2020) and (Boudrioua et al. 2020) | Normal (8,851)COVID-19 (5 9 0)Pneumonia (6,057) |
DenseNet | 92.00% |
| Ozcan, 2021 | X-Ray Image Dataset (Ozturk et al. 2020) | Normal (5 0 0)COVID-19 (1 2 5)Pneumonia (5 0 0) |
AlexNet + ResNet50 | 2-Class: 99.52% 3-Class: 87.64% |
| Irfan et al. 2021 | GitHub, COVID-19 radiography database, Kaggle, COVID-19 image data collection, and COVID-19 Chest X-ray Dataset | X-ray ImagesNormal (1100)COVID-19 (1900)Pneumonia (2000) CT ImagesNormal (6 0 0)COVID-19 (7 0 0)Pneumonia (1000) |
Hybrid deep neural networks (HDNN) consist of dropout, convolution, max-pooling layer, LSTM blocks, and a fully connected layer | 3-Class: 99% |
| Almalki et al. 2021 | COVID-19 Chest X-ray Dataset, Kaggle repository “Chest X-Ray Images, | A total of 1251 images were taken from the repositoriesNormal (620 samples)Pneumonia (660 samples)Viral-pneumonia (654 samples)Corona (568 samples) |
CoVIR-net Model (Inception + Resnet Models) |
CoVIR-net + Random Forest Multi-class: 97.29% |
COVID-19 X-ray image: CO19-Ximage Chest X-Ray Images: Ch-Ximage ChestX-ray8: Ch-X8image.