No. Papers Data Types of images AI methods to establish the algorithm CNN architecture Results for detecting COVID 1 [44 , 45 ] Total images: 4,356, COVID-19 images: 1,296, pneumonia images: 1,735, nonpneumonia images: 1,325 CT 3D deep learning ResNet-50 and COVNet Area under the curve (AUC): 0.96 2 [44 , 46 ] Total images: 618, COVID-19 images: 219, influenza-A (H1N1, H3N2, H5N1, H7N9, and others), images: 224, normal healthy lungs images: 175 CT 3D CNN model for segmentation Location-attention network and ResNet-18 Accuracy of 86.7%, average time: 30 s 3 [44 , 47 ] (PA) posterior-anterior images: 5,941, normal images: 1,583, bacterial pneumonia images: 2,786, non-COVID-19 viral pneumonia images: 1,804, COVID-19 images: 68 CXR Drop weights based Bayesian CNNs Bayesian ResNet50V2 Accuracy of 89.92% 4 [44 , 48 ] COVID-19 images: 453, training images: 217 CT Inception migration-learning model Internal validation: accuracy: 82.9%, specificity: 80.5%, sensitivity: 84%; External testing dataset: accuracy: 73.1%, specificity: 67%, sensitivity: 74% 5 [44 , 48 ] Total images: 1,065, COVID-19 images: 325; viral pneumonia images: 740 CT Modified inception transfer-learning model Accuracy: 79.30%, specificity: 0.83, sensitive: 0.67 6 [44 , 49 ] Total patients: 133, severe/critical patients: 54, nonsevere/critical patients: 79 CT Multilayer perception and long short term memory (LSTM) Area under the curve (AUC): 0.954 7 [44 , 50 ] Total images: 4,266, COVID-19 images: 2,529, CAP images: 1,338, influenza A/B images: 135, standard images: 258, total patients: 3,177, COVID-19 patients: 1,502, influenza A/B patients: 83, CAP patients: 1,334, healthy subjects: 258 CT 2D deep learning CNN ResNet 152 Accuracy: 94.98%, AUC 97.71%, sensitivity: 90.19%, specificity: 95.76%, the average time is taken to read: 2.73 s 8 [44 , 51 ] Total 1,136 cases from 5 hospitals, COVID-19 images: 723, non-COVID-19 images: 413 CT 3D deep learning method UNet ++ & ResNet-50 Specificity: 0.922, sensitive: 0.974 9 [44 , 52 ] , COVID-19 patients: 50, ordinary people: 50, CXR 5 pretrained CNN ResNet-50, ResNet101, ResNet52, InceptionV3, and inception-ResNetV2 ResNet-50: accuracy: 98.0% 10 [44 , 53 ] Total images:13,975, total patients:13,870 CXR Deep learning CNN COVID-net Accuracy: 92.4% 11 [44 , 54 ] Total patients: 157 CT CNN ResNet-50 Area under the curve (AUC): 0.996 12 [34 , 44 ] Normal images: 1,341, viral pneumonia images: 1,345, COVID-19 images: 190 CXR CNN AlexNet, ResNet-18, DenseNet-201, SqueezeNet Accuracy: 98.3% 13 [44 , 55 ] Total COVID-19 images: 531, CXR images: 170, CT images: 361 CT and CXR CNN with transfer learning Pretrained AlexNet Accuracy: CXR images: 98.3%, CT image: 94.1% 14 [6 ] Total images: 5,232, normal images: 1,346, bacterial pneumonia images: 2,538, viral pneumonia images: 1,345 CXR Deep learning framework using transfer learning Pretrained on ImageNet, trained using AlexNet, ResNet18, inception V3, DenseNet121, GoogLeNet, and ensemble model Ensemble model: accuracy: 96.4%, recall: 99.62% (unseen data) 15 [5 ] Total images: 5,247, bacterial pneumonia images: 2,561, viral pneumonia images 1,345, normal images: 1,341 CXR Pretrained deep CNN and used for transfer learning AlexNet, ResNet18, DenseNet201, and SqueezeNet DenseNet201 accuracy: normal and pneumonia: 98%, normal images, bacterial, and viral pneumonia: 93.3%, bacterial and viral pneumonia: 95% 16 [17 ] Total images: 306, COVID-19 images: 69, normal images: 79, bacterial pneumonia images: 79, viral pneumonia images: 79. The dataset number increases to 8,100 images after using the GAN network. CXR Deep transfer learning: using GAN network to generate more images to help detect the virus. Three deep transfer models. AlexNet, GoogLeNet, Restnet18 with performance measures in different scenario and classes GoogLeNet accuracy: 80.56% 17 [56 ] Dataset was collected from medRix and bioRxiv; COVID-19 images: 349, total patients: 216 CT Multitask learning and self-supervised DenseNet-169, ResNet-50 F1 score: 0.90, AUC: 0.98, accuracy: 0.89 18 [36 ] Total images: 2,200, COVID-19 images: 800, viral pneumonia images: 600 CT Machine learning technique using Microsoft Azure ResNet High accuracy: 91%, overall accuracy: 87.6% 19 [57 ] Total images: 15,495, normal images: 12,544, COVID-19 image: 2,951 CXR CNN model UNet, UNet++, DLA, DenseNet-121, CheXNet; inception-v3, ResNet-50 F1 score: 85.81%, sensitivity: 98.37%, specificity: 99.16% 20 [58 ] Diverse datasets from a different source CT Deep fully convolutional networks (FCN) UNet, ResDense FCN DSC: 0.780, sensitivity: 0.822, specificity: 0.951 21 [59 ] Total images: 954, COVID-19 images: 308, normal images: 323, pneumonia images: 323 images CXR Deep learning modules using stacked architecture concept DenseNet; GoogleNet Sensitivity: 0.91, specificity: 0.95, F1 score: 0.91, AUC: 0.97 22 [52 ] Total images: 7,406, COVID-19 images: 341, normal images: 2,800, viral pneumonia images: 1,493, bacterial pneumonia images: 2,772 CXR 2D five pretrained CNN based models ResNet50, ResNet101, ResNet152, InceptionV3, and inception-ResNetV2 COVID-19 and normal: accuracy: 96.1%, COVID-19 and pneumonia accuracy: 99.5%, COVID-19 and bacterial accuracy: 99.7% 23 [60 ] Total images (COVID-19, pneumonia, and normal): 1,266, COVID-19 images: 924 CT 3D pretrained the deep learning system and validate it. DNN Sensitivity (train): 78.93%, specificity (train): 89.93%, sensitivity (val): 80.39%, specificity (val): 81.16% 24 [61 ] Total images (COVID-19, bacterial, and normal): 275, COVID-19 images: 88 CT 2D pretrained ResNet 50 using the feature pyramid network (FPN) DRE-net Sensitivity: 93%, specificity: 96%, accuracy: 99% 25 [62 ] Total images: 624, COVID-19 images: 50 CXR 2D GAN + TL AlexNet, GoogLeNet, ResNet18, SqueezeNet Accuracy: 99% 26 [63 ] Total images (COVID-19, bacterial, and normal): 1,427, COVID-19 images: 224, bacterial and viral pneumonia images: 714 CXR 2D transfer learning (TL) VGG19, MobileNet, Inception, Xception, Inception ResNet v2. Sensitivity: 98.66%, specificity: 96.46%, accuracy: 94.72% 27 [64 ] Total images (COVID-19, pneumonia, normal): 6,008, COVID-19 images: 184 CXR 2D transfer learning (TL) Three ResNet models Accuracy: 93.9% 28 [65 ] Total images (COVID-19, pneumonia, and normal): 8,850, COVID-19 images: 498 CXR 2D convolutional autoencoder (CAE) AE: COVIDomaly Accuracy: 76.52% 29 [66 ] Total images (COVID-19, pneumonia, and normal): 2,905, COVID-19 images: 219 CXR 2D CNN + k-NN + SVM Accuracy: 98.70% 30 [67 ] Total images (COVID-19, pneumonia, and normal): 2,905, COVID-19 images: 219 CXR 2D using hyperparameters Bayesian optimisation algorithm ANN + AlexNet Sensitivity: 89.39%, specificity: 99.75%, accuracy: 98.97%, F-score: 96.72% 31 [68 ] Total images (COVID-19, pneumonia, and normal): 502, COVID-19 images: 180 CXR 2D patch-based convolutional neural network ResNet-18 Sensitivity: 76.90%, specificity: 100.00% 32 [69 ] Total images (COVID-19, pneumonia, and normal): 2,905, COVID-19 images: 219 CXR 2D Ensemble: Resnet50 and VGG16 Sensitivity: 91.24%, specificity: 99.82% 33 [70 ] Total images (COVID-19 and normal): 2,492, COVID-19 images: 1,262 CT 2D TL and DenseNet201 Accuracy: 99.82% 34 [71 ] COVID-19, pneumonia, and normal images from Cohen et al. [37 ] CXR 2D Xception Sensitivity: 97.09%, specificity: 97.29%, accuracy: 97.40% 35 [72 ] Total images (COVID-19 and normal): 380, COVID-19: 180 CXR 2D 5 pretrained models+ SVM Accuracy: 94.7% 36 [73 ] Total images (COVID-19, pneumonia, normal, and non-COVID-19): 2,905, COVID-19 images: 219 CXR 2D pretrained models such as ResNet101, Xception, InceptionV3, MobileNet, and NASNet InstaCovNet-19 Accuracy: 99.08%, accuracy: 99.53% 37 [74 ] Datasets contain COVID-19, pneumonia and normal images. CXR 2D 5 pretrained CNNs Accuracy: 95.00% 38 [75 ] Datasets contain bacterial pneumonia, non-COVID viral pneumonia, and COVID-19 images. CXR 2D 5 COVID-CAPS Sensitivity: 90%, specificity: 95.8%, accuracy: 95.7% 39 [76 ] Total images (COVID-19 and normal): 5,000, COVID-19 images: 184 CXR 2D 5 TL + pretrained models Sensitivity: 100%, specificity: 98.38% 40 [55 ] Total images (COVID-19 and normal): 526, COVID-19 images: 238 CXR + CT 2D TL + AlexNet model Sensitivity: 72%, specificity: 100%, accuracy: 94.1% 41 [77 ] Total images (COVID-19 and normal): 320, COVID-19 images: 160 CXR + CT 2D Apache spark framework TL + inceptionV3 & ResNet5 Sensitivity: 72%, specificity: 100%, accuracy: 99.01% 42 [78 ] Total images (COVID-19, pneumonia, and normal): 4,575, COVID-19 images: 1,525 CXR 2D CNN used for deep feature extraction, and LSTM is used for detection using the extracted feature LSTM+CNN Sensitivity: 99.2%, specificity: 99.9%, accuracy: 99.4% 43 [79 ] Dataset 1 images (COVID-19, pneumonia, and normal): 4,448, COVID-19 images: 2,479, dataset 2 images (COVID-19, pneumonia, and normal): 101, COVID-19 images: 52 CXR 2D 3D inception V1 Dataset 1: accuracy: 99.4%; dataset 2: sensitivity: 98.08%, specificity: 91.30%, accuracy: 93.3% 44 [80 ] Total images (COVID-19, pneumonia, and normal): 1,343, COVID-19 images: 446 CXR 2D Conditional GAN: LightCovidNet Accuracy: 97.28% 45 [81 ] Total images (COVID-19 and normal): 8,504, COVID-19 images: 445 CXR 2D TL VGG-16 model Sensitivity: 98.0%, specificity: 100.00%, accuracy: 94.5% 46 [82 ] Total images (COVID-19 and normal): 746, COVID-19 images: 349 CT 2D TL+ ensemble of 15 pretrained models: EfficientNets(B0-B5), NasNetLarge, NasNetMobile, InceptionV3, ResNet-50, SeResnet 50, Xception, DenseNet121, ResNext50, and Inception_resnet_v2 Accuracy: 85.0% 47 [83 ] Total images (COVID-19 and normal): 2,482, COVID-19 images: 1,252 CT 2D AE + random forest Specificity: 98.77%, accuracy: 97.87% 48 [84 ] Total images (COVID-19 and normal1): 50, COVID-19 images: 25 CXR 3D COVIDX-net Sensitivity: 100.00%, specificity: 80.00% 49 [85 ] Total images (COVID-19 and Normal): 800, COVID-19 images: 400 CXR 2D using modern and traditional machine learning methods: (ANN), (SVM), linear kernel and (RBF), - nearest neighbor ( - NN), decision tree (DT), and CN 2 rule inducer techniques Deep learning models: MobileNets V2, ResNet50, GoogleNet, DarkNet, and Xception ResNet50 accuracy: 98.8% 50 [86 ] Total images (COVID-19 and Normal): 800, COVID-19 images: 400 CXR 2D CLAHE and Butterworth bandpass filter was applied to enhance the contrast and eliminate the noise. The hybrid multimodal deep learning system COVID-deep net system. Sensitivity: 99.9%, specificity: 100.0%, accuracy: 99.3% 51 [87 ] Datasets from Cohen et al. [37 ]. Total images (COVID-19 and normal): 800, COVID-19 images: 400 CXR 2D benchmarking and diagnostic models: decision matrix that embedded a mix of 10 evaluation criteria and 12 diagnostic models, also known as multicriteria decision making (MCDM) TOPSIS is applied for benchmarking and ranking purpose, while entropy is used to calculate the criteria’s weights. SVM is selected as the best diagnosis model Coefficient value: 0.9899 52 [88 ] Total images (COVID-19 and normal): 800, COVID-19 images: 400 CXR 2D hybrid deep learning framework, pretrained deep learning models incorporating of a ResNet34, and high-resolution network model COVID-CheXNet system Sensitivity: 99.98%, specificity: 100.0%, accuracy: 99.99%