|
| Reference (country) | Aim of the study | Population | Feature engineering | ML method | Model | Type of data | Validation results |
|
| Saiz and Barandiaran, [62] (Spain) | Detection | 1500 | Automatic feature extraction | CNN with transfer learning | VGG-16 SDD | X-ray | Accuracy: 94.92%, sensitivity: 94.92%, specificity: 92% |
| Ni et al., [34] (China) | Detection | 14531 | Prominent features selected | Deep learning | Convolutional MVP-Net and 3D U-Net | CT images | F1 score: 97% in detecting lesions, sensitivity: 100%, for detecting patient sensitivity for per-lung Lobe lesion: 0.96% |
| Wang et al., [21] (China) | Diagnosis and prognosis | 5372 (two datasets) | Not used for diagnosis | Deep learning | DenseNet121-FPN | CT images | AUC = 87% and 88%, sensitivity: 80.3% and 79.35%, specificity: 76.3% and 81.1% |
| Rahimzadeh & Attar, [50] (Iran) | Diagnosis | 11302 images (open source) | Automatic feature extraction | Deep learning | Xception and ResNet50V2 | X-ray | Accuracy: 95.5%, overall average accuracy: 91.4% |
| Panwar et al., [36] (India) | Fast detection | 337 images (open source) | Not used | Deep learning (nCOVnet) | VGG-16 | X-ray | Sensitivity: 97.62%, specificity: 78.57%, accuracy: 88.10% |
| Ardakani et al., [30] (Iran) | Detection | 194 | Not used | Deep learning | AlexNet, VGG-16, VGG-19, SqueezeNet, GoogLeNet, etc. | CT images | Sensitivity: 100%, specificity: 99.02%, accuracy: 99.51% |
| Li et al., [46] (China) | Diagnosis | 4356 CT exams from 3322 patients | Automatic feature extraction | Deep learning | ResNet-50 as backbone of main model | CT images | Sensitivity: 90%, specificity: 96% |
| Li et al., [19] (Greece) | Automatic diagnosis | 2914 | Automatic feature extraction | CNN with transfer learning | MobileNetV2 | X-ray | Accuracy: 96.78%, sensitivity: 98.66%, specificity: 96.46% |
| Sethy et al., [51] (India) | Diagnosis | 381 | Automatic feature extraction | CNN and SVM | ResNet-50 | X-ray | Sensitivity: 95.33% |
| Song et al., [37] (Chain and USA) | Detection | 227 | Automatic feature extraction | Deep learning (CoroNet) | BigBiGAN1 | CT images | Sensitivity: 85%, specificity: 88% |
| Brunese et al., [31] (Italy) | Detection | 6,523 | Automatic feature extraction | Deep learning (CoroNet) | VGG-16 | X-ray | Accuracy: 97% |
| Butt et al., [42] (USA) | Classification (diagnosis) | 618 | Automatic feature extraction | CNN | ResNet-18 | CT images | Sensitivity: 98.2%, specificity: 92.2% |
| Loey and et al., [63] (Egypt) | Diagnosis (classification) | 306 | Automatic feature extraction | Deep learning | GoogLeNet | X-ray | Accuracy: 100% |
| Ozturk et al., [35] (Turkey) | Automated detection | 2 databases | Automatic feature extraction | Deep learning | DarkNet | X-ray | Binary case accuracy: 98.08%, multiclass cases accuracy: 87.02% |
| El Asnaoui and Chawki, [58] (Morocco) | Diagnosis | 6087 | Automatic | Deep learning | Inception_ResNet_V2 | X-ray and CT | Inception_ResNet_V2 accuracy: 92.18%, DenseNet201 accuracy: 88.09% |
| Yang et al., [59] (China) | Detection | 295 | Automatic | Deep learning | DenseNet | CT images | Accuracy: 92%, sensitivities: 97%, specificity: 0.87 |
| Jaiswal et al., [33] (India) | Detection | 2492 (open source) | Automatic | Deep transfer learning | DenseNet201 | CT images | Precision: 96.29%, specificity: 96.21%, accuracy: 96.25% |
| Mahmud et al., [61] (Bangladesh) | Diagnosis | 5856 | Not mentioned | Deep learning (CNN) | CovXNet | X-ray | Accuracy of multiple classes: 90.2% |
| Singh et al., [52] (India) | Classification (diagnosis) | Not mentioned | Automatics using CNN | CNN, ANN, and ANFIS | Not mentioned | CT images | Proposed model is compared with CNN, ANFIS, and ANN models and it shows high performance |
| Ko et al., [45] (Korea) | Diagnosis (differentiate) | 3993 patients | Automatic feature extraction | Deep learning (FCONet) | ResNet-50 | CT images | Sensitivity: 99.58%, specificity: 100.00%, accuracy: 99.87% |
| Wu et al., [56] (China) | Screening (diagnosis) | 495 | Automatic feature extraction | Deep learning (CoroNet) | VGG-19 | CT images | Accuracy: 76.0%, sensitivity: 81.1%, specificity: 61.15% |
| Vaid et al., [38] (Canada) | Detection | 181 | Automatic feature extraction | Deep learning (CoroNet) | VGG-19 | X-ray | Accuracy: 96.3% |
| Ucar & Korkmaz, [54] (Turkey) | Classification (diagnosis) | Public | Automatic feature extraction | CNN | Deep Bayes SqueezeNet | X-ray | Accuracy for overall class: 98.3% |
| Toğaçar et al, [53] (Turkey) | Diagnosis | Two open sources (n = 295) | Automatic feature extraction | Deep learning (CoroNet) | SqueezeNet and MobileNet | X-ray | Classification rate: 99.27% |
| Khan et al., [57] (India) | Detection and diagnosis | Two datasets (n = 1300) | Automatic feature extraction | Deep learning (CoroNet) | Xception | X-ray | Accuracy: 89.6% |
| Wu et al., [56] (China) | Screening (diagnosis) | 495 | Automatic feature extraction | Deep learning (CNN) | ResNet-50 | CT images | Accuracy: 0.819%, sensitivity: 0.760%, specificity: 0.811% |
| Yi et al., [40] (USA) | Classification (detection) | 88 | Automatic feature extraction | Deep learning (CNN) | Not mentioned | X-ray | Sensitivity: 89% |
| Martínez et al., [64] (Columbia) | Detection | 240 | Automatic feature extraction | CNN | NASNet2 | X-ray | Accuracy: 97% |
| Das et al., [43] (India) | Screening (diagnosis) | 6845 | Automatic feature extraction | Deep learning (CNN) | Truncated inception net | X-ray | Sensitivity: 88%, specificity: 100% |
| Hasan et al., [44] (Iraq) | Diagnosis (classification) | 321 | Q-deformed entropy feature extraction | Deep transfer learning | LSTM neural network classifier | CT images | Accuracy: 99.68% |
| Pathak et al., [65] (India) | Classification (detection) | 852 | Automatic feature extraction | Transfer learning technique | ResNet-50 | CT images | Accuracy: 93.01% |
| Waheed et al., [39] (India) | Detection | 1124 | Automatic feature extraction | GAN (CovidGAN) | ACGAN3, VGG-16 | X-ray | Accuracy: 95%, sensitivity: 90%, specificity: 97% |
| Pereira et al., [49] (Brazil) | Diagnosis (classification) | 1144 | Automatic feature extraction | Deep learning (CNN) | Inception-V3 | X-ray | F1 score: 89% |
| Mei et al., [48] (USA) | Diagnosis | 905 | Automatic feature extraction | Deep learning (CoroNet) | Inception_ResNet_V2 | CT images | Correctly identifying 17 of 25 (68%) patients, whereas radiologists classified all of these patients as COVID-19 negative |
| Brunese et al., [31] (Italy) | Detection and diagnosis | 6523 | Automatic feature extraction | Deep learning (CoroNet) | VGG-16 | X-ray | Accuracy: 96.3% |
| Apostolopoulos et al., [29] (Greece) | Detection | 455 | Automatic feature extraction | Deep learning (CoroNet) | MobileNetV2 | X-ray | Sensitivity: 97.36%, specificity: 99.42%, accuracy: 99.18% |
| Elaziz et al., [60] (Egypt) | Detection | 2 databases (open source) | FrMEMs4 | Deep learning (CoroNet) | MobileNet | X-ray | Accuracy for first dataset: 96.09%, accuracy for second dataset: 98.09% |
|
