Skip to main content
. 2021 Dec 1;21(23):8045. doi: 10.3390/s21238045

Table 2.

State-of-the-art AI techniques to detect COVID-19 using chest X-ray imagery.

Paper Method Used: Preprocessing + Segmentation + Feature Extraction + Feature Selection + Classification or CNN + Classification Result Obtained Dataset Used
(Most Are Public)		 No. of Classes
[114] Image enhancement + WS +deep CNN (ResNet50) and DWT and GLCM+ mRMR+ RF Cvd.Acc: 99.45, Cvd.Sen:.99.17, Cvd.Pre: 97.51,F1-Score: 0.9833 N:1500,C-19: 790,BP: 1304,VP: 1215
(after data augmentation)		 2 (C-19, NC)
Cvd.Acc: 98.48, Cvd.Sen: 98.72, Cvd.Pre: 97.89,F1-Score: 0.9829 4
[115] Color layout descriptor + k-NN Cvd.Sen: 96.5, Cvd.Pre: 96.5 Total:86
[116] CNN model + Long short-term memory (LSTM) Cvd.Acc: 99.4, Cvd.Sen: 99.3, Cvd.Spe: 99.2, F1-Score: 98.9, AUC: 99.9 N: 1525, C-19: 1525,P: 1525 3
[117] Concatenation of the Xception and ResNet50V2 Cvd.Acc (avg.): 91.4 N: 8851,C-19: 180,P: 6054 3
[118] CNN model Cvd.Acc: 95, Cvd.Sen: 96.9, Cvd.Spe: 97.5, Cvd.Pre: 95, F-measure: 95.6 N: 310,C-19: 284,BP: 330,VP: 327 3(N, C-19, P)
Cvd.Acc: 89.6, Cvd.Sen: 89.92, Cvd.Spe: 96.4, Cvd.Pre: 90,F-measure: 96.4 4
[119] CNN model AUROC: 0.96 Pvt. + Public Dataset 3
[120] DarkNet based CNN model Cvd.Acc(avg.): 98.08, Cvd.Sen(avg.): 95.13, Cvd.Spe(avg.): 95.3, Cvd.Pre (avg.): 98.03,F1-Score (avg.): 96.51 N: 500,C-19: 127,P: 500 2 (N, C-19)
Cvd.Acc(avg.): 87.02, Cvd.Sen(avg.): 85.35, Cvd.Spe(avg.): 92.18, Cvd.Pre (avg.): 89.96,F1-Score (avg.): 87.37 3
[121] 2D-CTf + CSSA+ EfficientNet-B0 Cvd.Acc: 99.69, Cvd.Sen: 99.44, Cvd.Spe: 99.81, Cvd.Pre: 99.62, F-measure: 99.53 N: 1281,C-19: 159,VP: 1285 3
[122] VGG-16 model Cvd.Acc(avg.): 97 N: 3520,C-19: 250,P: 2753 3
[123] ResNet50 + ResNet101 Cvd.Acc: 97.77, Cvd.Sen: 97.14, Cvd.Pre: 97.14 N: 315,C-19: 250, BP: 300,VP: 350 2(C-19,O)
[58] ResExLBP + Relief-F+ SVM Cvd.Acc: 99.69, Cvd.Sen: 98.85, Cvd.Spe: 100 N: 234, C-19: 87 2
[124] VGG16 model Cvd.Acc: 98.1 N: 2880, C-19: 415, P: 5179 2(C-19,NC)
Cvd.Acc: 94.5 3
[125] ResNet18, ResNet50, SqueezeNet,& DenseNet121 Cvd.Sen: 98, Cvd.Spe(avg.): 90 C-19: 200, NC:5000 2
[126] Capsule Network-based architecture Cvd.Acc: 95.7, Cvd.Sen: 90, Cvd.Spe: 95.8, AUC: 0.97 2(C-19,O)
[127] VGG16 model Cvd.Sen: 97.62, Cvd.Spe: 78.57 N:142, C-19: 142 2
[128] ResNet101 Cvd.Acc: 71.9, Cvd.Sen: 77.3, Cvd.Spe: 71.8 C-19: 154, NC: 5828 (test data) 2
[129] Deep learning model Cvd. Acc C-19: 100,P: 93.75,N: 100 N: 66, C-19: 51,NC: 21,P: 160,TB: 54 5
[130] Sequential CNN model Cvd.Acc: 98.3, Cvd.Sen: 100, Cvd.Pre: 96.72, F1-Score: 98.3,ROC area: 0.983 N: 659, C-19: 295 2
[131] HE +VGG16-based model Cvd.Acc (avg.): 86, Cvd.Sen (avg.): 86, Cvd.Spe(avg.): 93, Cvd.Pre(avg.):86,F1-Score: 86 N: 132, C-19: 132,P: 132 3
[132] Histogram matching and autoencoder and CLAHE + Custom CNN model Cvd.Acc (avg.):94.43, Cvd.Sen (avg.): 92.53, Cvd.Spe: 96.33, Cvd.Pre(avg.): 93.76,
F1-Score (avg.): 93.14,AUC (avg): 0.988		 N: 4337,C-19: 2589 2
[133] Ensemble of ResNet-18 Model Cvd.Acc: 95.5, Cvd.Sen: 100, Cvd.Pre: 94 N: 1579,C-19: 184,P: 4245 3
[134] HE+ lung segmentation using UNet + Various deep model are analyzed.
[135] 4 models analyzed (Best: VGG16 and VGG19) Cvd.Acc: 99.38, Cvd.Sen: 100, Cvd.Spe: 99.33 N: 802, C-19: 790 2
[136] CLAHE+VGG16 and VGG19 used (Best: VGG16) Cvd.Acc: 95.9, Cvd.Sen: 92.5, Cvd.Spe: 97.5,AUC: 0.950 (max. only for C-19) N: 607,C-19: 607,P: 607 3
[137] CNN model to separate COVID-19 and pneumonia
[138] Alexnet, Googlenet, and Restnet18 is used
(Googlenet best for 4 classes)		 Cvd.Acc: 80.56, Cvd.Sen: 80.56, Cvd.Pre: 84.17, F1-Score: 82.32 N: 79,C-19: 69, BP: 79, VP: 79 4
[76] MLP-CNN Cvd.Acc: 95.4, Cvd.Sen: 95, Cvd.Pre: 92.5, F1-Score: 93.6 C-19: 112, NC: 30 2
[139] LightCovidNet Cvd.Acc (avg.): 96.97 N: 1341,C-19: 446,P: 1345 3
[140] MobileNet v2 Cvd.Acc: 96.78, Cvd.Sen: 98.66, Cvd.Spe: 96.46 N: 504, C-19: 224, P: 714 2(C-19,O)
Cvd.Acc: 94.72 3(N,C-19,P)
[141] Truncated InceptionNet Cvd.Acc (avg.): 98.77, Cvd.Sen(avg.): 95, Cvd.Spe(avg.): 99, Cvd. Pre(avg.): 99
F1 score(avg.): 0.97, AUC (avg.):0.99		 N:2003, C-19:162,P: 4280, TB:400
 4
[142] CNN model Cvd. Prec (avg.), Cvd. Sen (avg.), F1-score (avg.): 100 C-19: 500, P: 500 2
[143] CNN model Cvd.Acc (testing): 94.4 N:8066, C-19:183,P: 5551 3
[144] COVID-Net model Cvd.Acc: 93.3 Total: 13,975 from 13,870 patients 3(N,C-19,P)
[85] CNN model (Inception) + FO-MPA + k-NN Cvd.Acc: 98.7, F-score: 98.2 DS1: C-19 +ve: 200, C-19 -ve: 1675 2
Cvd.Acc: 99.6, F-score: 99 DS2: C-19 +ve: 219, C-19 -ve: 1341
[63] FrMEMs + MRFO + k-NN Cvd.Acc: 96.09, Cvd.Sen: 98.75, Cvd.Pre: 98.75 DS1: C-19 +ve: 216,C-19 -ve: 1675 2
Cvd.Acc: 98.09, Cvd.Sen: 98.91, Cvd.Pre: 98.91 DS2: C-19 +ve: 219,C-19 -ve: 1341
[145] Xception model + SVM Cvd.Acc: 99.33, Cvd.Sen: 99.27, Cvd.Spe: 99.38, Cvd.Pre: 99.27, F1-score:99.27,AUC: 99.32 N: 565,C-19: 537 2
[146] Discriminative cost sensitive learning approach Cvd.Acc: 97.01, Cvd.Pre: 97, Cvd.Sen: 97.09,F1-score: 96.98 N: 1000,C-19: 239,P: 1000 3
[147] CNN model Cvd.Sen (avg.): 91.05, Cvd.Spe(avg.): 99.61, Cvd.Acc(avg.): 98.34,ROC-AUC(avg.): 95.33 N: 1583,C-19: 225 2
Cvd.Sen (avg.): 92.88, Cvd.Spe(avg.): 99.79, Cvd.Acc(avg.): 99.44,ROC-AUC(avg.): 96.33 C-19: 225, P: 4292 2
F1 score (avg.): 94.10 N: 1583,C-19: 225,P: 4292 3
[148] HE and GC + DenseNet103 + ResNet18
 Cvd.Acc: 91.9 N: 191, C-19: 180,BP: 54, VP: 20,TB: 57 4(N,BP,VP,TB)
[149] VGG16 model Cvd.Acc, Cvd.Sen, Cvd. Prec, F-score: 80 C-19: 70, NC: 70 2
[54] ACGAN based model (CovidGAN) Cvd.Acc: 95.00 N: 403, C-19: 721 2(N, C-19)
[150] CNN model Cvd.Acc: 99.70, Cvd.Pre: 99.70, Cvd.Sen: 99.70, Cvd.Spe: 99.55 N: 1579, C-19: 423,VP:1485 2(N,C-19VP)
[151] Deep learning model Cvd.Acc: 97.25, Cvd.Pre: 97.24,F1-score: 97.21 N: 27,228, C-19: 209, P: 5794 3
[152] CNN + gated recurrent unit (GRU) Cvd.Sen: 96, Cvd.Pre: 96, F1-score: 95 N: 141, C-19: 142, P: 141 3
[153] Ensemble of deep CNN model (InceptionResNetV2 + ResNet152V2 + VGG16+ DenseNet201) Cvd.Acc: 99.2, Cvd.Sen: 99.12, Cvd.Spe: 99.07, F-score: 99.17,AUC: 99.21 N:2039, C-19:1663,P: 401,TB:394 4
[154] MCFF-Net66-Conv1-GAP Cvd.Acc: 94.66 N:1500,C-19:942, BP:1802,VP:1797 4
[155] ResNet50V2 + t-SNE Cvd.Acc: 95.49, Cvd.Sen: 99.19, Cvd.Pre:96.19, F1-score: 98.0, AUC: 95.49 N: 616, C-19: 616,P: 616 3
[156] CNN model Cvd.Acc:100, Cvd.Sen:100, Cvd.Spe:100, Cvd.Prec:100, F1-score:100, AUC:100 N:42, C-19:136 2
[157] Enhanced Inception-ResNetV2 model Cvd.Acc(avg.): 98.80, Cvd.Sen(avg.): 99.11, Cvd.Prec(avg.): 98.61,F1 score(avg.): 98.86 N:1341,C-19:219,VP: 1345 3
[158] CNN model and GoogLeNet Cvd.Acc: 97.62, Cvd.Sen: 98.29, Cvd.Spe: 97.64, F-score: 98.30,AUC: 97.96 N: 1421,C-19: 1332 2
[159] VGG16 Model Cvd.Acc: 98.72, Cvd.Sen: 98.78, Cvd.Spe: 98.70, Cvd.Prec: 96.43, F1-score: 97.59 N:1341,C-19:1200,VP:1345 3
[160] AlexNet Cvd.Acc: 99.13, Cvd.Sen: 99.4, Cvd.Spe: 99.15,F-score: 99.49,AUC: 99.31 Consists: N,C-19,P,TB 4
[161] Ensemble of MobileNet and InceptionV3 Cvd.Acc: 96.49, Cvd.Prec: 93.01, Cvd.Sen: 92.97,F-score: 92.97 N:1050,C-19:1050,BP:1050,VP:1050 4
[162] VGG16 model Cvd.Acc(avg.): 91.69, Cvd.Sen(avg): 95.92, Cvd.Spe(avg.): 100 Total: 7720 3(N, C-19,P)
[163] CLAHE + InceptionV3 + ANN Cvd.Acc: 97.19 N: 1583,P: 4273 2
[97] CNN with various optimization algorithm Cvd.Acc:96, Cvd.Sen:100, Cvd.Spe:99, Cvd.Pre:96, F1-Score:0.98 N: 1583, C-19: 576, VP:4273 3
[164] VGG16 model Cvd.Acc: 96, Cvd.Sen: 92.64, Cvd.Spe: 97.27 N: 504, C-19: 224 2
Cvd.Acc: 92.53, Cvd.Sen: 86.7, Cvd.Spe: 95.1 N:504, C-19: 224, P: 700 3
[50] FOSF and GLCM and HOG + GWO + Ensemble of classifiers Cvd.Acc: 98.06, Cvd.Sen: 98.83, Cvd.Spe: 96.51, Cvd.Pre: 98.26,F-measure: 98.55 AUC:0.97 N: 782, C-19: 782, P: 782 2 (N,AB)
Cvd.Acc: 91.32, Cvd.Sen: 96.51, Cvd.Spe: 86.2, Cvd.Pre:87.36,F-measure: 91.71,AUC: 0.91 2(C-19,P)
[165] Ensemble of deep CNN model (VGG19 + DenseNet121) + SVM Cvd.Acc: 99.71 N:2341, C-19: 798,P: 2345 2 (C-19,NC)
Cvd.Acc: 98.28, Cvd.Sen (avg), Cvd.Pre(avg.),F1-Score (avg.): 98.33 3
[166] CNN model + Ensemble of classifiers Cvd.Acc: 98.91, Cvd.Sen: 97.82, Cvd.Pre: 100,F1-Score: 98.89 N: 2300,C-19: 2300 2
[167] Deep learning model (Inception architecture) Cvd.Acc: 96, Cvd.Sen: 93, Cvd.Spe: 97, Cvd.Pre: 97, F1-Score: 0.96 C-19: 435,NC: 505 2
[168] UNet with ResNet + CNN model Cvd.Acc (avg.): 96.32 N:1840,C-19:433,BP:2780,VP:1345,TB: 394 5
[169] Two separate CNN models for binary and ternary classification Cvd.Acc: 98.7, Cvd.Sen: 100, Cvd.Spe: 98.3 N:145,C-19: 145, BP: 145 2(N, C-19)
Cvd.Acc: 98.3, Cvd.Sen: 99.3, Cvd.Spe: 98.1 3
[170] VGG16 and Xception model (Best: Xception) Cvd.Sen: 100, Cvd.Spe: 97.6, F1-Score: 97.7 N: 400, C-19: 402,P:200,I: 35 2
[171] Various DNN + Majority voting scheme Cvd.Acc: 99.31 N: 1338, C-19: 237, VP: 1336 3
[172] Customized CNN Model Cvd.Acc: 92.95, Cvd.Sen (avg.): 90.72, Cvd.Pre(avg.): 94.04,F1-Score(avg.): 0.9204 N: 1341, C-19: 744 (Independent set) 2
[173] NanoChest-net model Analyzed with various datasets.
[174] VGG16+ HS + k-NN Cvd.Acc, Cvd.Sen, Cvd.Pre,F1-Score, AUC:100 N: 480,C-19: 280 2
[175] OptiDCNN model Cvd.Acc: 99.11 N: 5000, C-19: 184 2
[176] HOG and CNN(VGG19) + ME + CNN classifier + WS Cvd.Acc: 99.49, Cvd.Sen: 93.65, Cvd.Spe: 95.7 C-19 +ve: 1979, C-19 -ve: 3111 2
[177] Ensemble-CNNs (based on ResNeXt-50, Inception-v3, and DenseNet-161) Cvd.Acc: 75.23 ± 3.40, Cvd.Sen: 75.20, Cvd.Spe: 87.60, Cvd.Pre: 78.28, F1-Score: 73.43
AUC: 0.8140		 N: 711, C-19: 711,P:711,BP:711,VP:711
Lung Opacity not Pneumonia:711 (public+Pvt.)		 3(N,C-19,P)
Cvd.Acc: 81.00 ± 2.39, Cvd.Sen: 82.96, Cvd.Spe: 85.24, Cvd.Pre: 82.99,F1-Score: 81.49,
AUC: 0.8810		 5
[178] Showed that a system with 2-class model are not valid for the diseases with similar symptoms, by conducting various experiments
[179] Exemplar COVID-19FclNet9 + SVM Cvd.Acc: 99.64 N: 150,C-19:127 2
Cvd.Acc: 98.84 N: 4000,C-19: 3616, P: 1345 3
Cvd.Acc: 97.60 N: 234,C-19:125,BP:242,VP:148 4
[180] Decompose, Transfer, and Compose (DeTraC)+PCA Cvd.Acc: 93.1, Cvd.Sen:100 N: 80, C-19:105,SARS: 11 3
[77] UNet + HRNet Cvd.Acc: 99.26, Cvd.Sen:98.53, Cvd.Spe: 98.82 Total: 272 2
[181] Various CNN model used (Best:EfficientNetB0) Cvd.Acc:92.93, Cvd.Sen: 90, Cvd.Spe: 95, Cvd. Prec: 88.3,F1- score: 0.88 N: 1341, C-19: 420, P: 1345 3
[182] EfficientNet B3-X Cvd.Acc: 93.9, Cvd.Sen: 96.8, Cvd.PPV: 100 N:7966+100, C-19: 152+31 P: 5421+100 3
[183] Various pre-trained CNN models (Best: ResNet50) Cvd.Acc: 96.1 (N,C-19), Cvd.Acc: 99.5(C-19,VP), Cvd.Acc: 99.7(C-19,BP) N: 2800, C-19: 341, BP: 2772, VP: 1493 2
[184] CNN model + SVM Cvd.Acc (avg.): 95.81, Cvd. Prec(avg.): 95.27, F1 score(avg.): 94.94 N:1266 +317, C-19:460 + 116 P:3418 + 855 (Pvt.) 3
[185] ResNet50+ SVM Cvd.Sen:80, Cvd.Spe: 81, AUC: 0.81 Training and validation
C-19:250, NC:250		 Testing independent set
C-19:74,NC:36
(Pvt.)		 2
[186] VisionPro Deep Learning™ + COGNEX’s F-score: 95.3 (for segmented lung) N: 7966+100,C-19: 258+100
P: 5451+100		 3
[84] Pillow library + HSGO + SVM Cvd.Acc:99.65 C-19: 371, NC: 1341 2
[187] CNN model Cvd.Acc (avg.): 98.03, Cvd.Sen(avg.): 98.83, Cvd.Spe(avg.): 97 DS1:C-19: 217, NC: 1126
DS2:C-19: 2025, NC: 2025		 2
[188] AlexNet + Relief + SVM Cvd.Acc: 99.18 N:1583, C-19: 219, P:4290 3
[189] RGB to YUV and YUV to RGB + CNN Cvd.Acc: 84.76, Cvd.Sen: 98.99, Cvd.Spe: 92.19, F-score: 0.9389,AUC: 0.5948 N:28,C-19:78,P: 79(each for BP and VP) 4
[190] CNN model Cvd.Acc: 98.44 Total: 392, C-19: 196 2
[191] Deep CNN model Cvd.Acc(avg.): 91.62, AUC:91.71 C-19 +ve: 538, C-19 –ve: 468 2
[192] Deep CNN model Cvd.Acc(avg.):99.2, Cvd.Sen(avg.):99.2,F1- score: 0.992 N, C-19: 2484 (each)
N, C-19,P: 3829 (each)		 2
Cvd.Acc(avg.):95.2, Cvd.Sen(avg.):95.2,F1-score: 0.952 3
[193] MobileNetV2 Cvd.Acc: 92.91, Cvd.Pre: 92 N: 234, C-19: 390 2
[49] DenseNet201 model+ Quadratic SVM Cvd.Acc: 98.16, Cvd.Sen: 98.93, Cvd.Spe: 98.77 N: 2924, C-19: 683,P: 4272 3
[194] Cluster-based learning + Ensemble of classifiers Cvd.Acc (avg.):100 N:79,C-19: 69, BP:79, VP:79 2(N,C-19)
Cvd.Acc(avg.): 85.23 3(N,C-19,BP)
Cvd.Acc(avg.): 74.05 4
[195] Various deep CNN models are compared
(Best: XCeptionNet)		 F1-score: 0.97 N: 1345+238, C-19:490+ 86,P:3632+ 641
(Train + Test)		 3
[196] CNN model Cvd.Acc: 98.19 N: 10,456, C-19: 573, P: 11,673 (Pvt.) 2(C-19,P)
Cvd.Acc: 91.21 3
[197] Federated learning model Cvd.Acc: 98.72 N: 1266, C-19: 460,P: 3418 (Pvt.) 2(C-19,P)
Cvd.Acc: 95.96 3
[80] ResNet50 + ASSOA + MLP Cvd.Acc: 99.70 Total: 5863 2(C-19+ve, C-19-ve)
[198] Several CNN models are analyzed (Best: VGG16) Cvd.Acc: 91 N:1341, C-19:219,P:1345 3
[199] Semi-supervised open set domain adversarial network (SODA) Avg. AUC-ROC Score: 0.9006(C-19), 0.9082(P) With different domain target dataset
[200] VGG16 model Cvd.Acc: 97, Cvd.Sen: 99, Cvd.Spe: 99, Cvd.Pre: 97, F-score: 98 N:1400, C-19: 210, P: 1400 3
[201] CovFrameNet (deep learning architecture) Cvd.Acc: 100, Cvd.Sen: 85, Cvd.Spe: 100, Cvd.Pre: 85, F-score: 90, AUC: 50 Using two different dataset
[202] Self-supervised super sample decomposition for transfer learning (4S-DT) model Cvd.Acc: 97.54, Cvd.Sen: 97.88, Cvd.Spe: 97.15 DS1: N: 296, C-19: 388, SARS: 41 3(N, C-19, SARS)
Cvd.Acc: 99.80, Cvd.Sen: 99.70, Cvd.Spe: 100 DS2: N: 1583,C-19: 576,P: 4273 3 (N,C-19,P)
[203] VDI + Residual encoder + SVM Cvd.Acc: 93.60, Cvd.Sen: 88, Cvd.Pre: 100, F1-score: 93.60 C-19: 315, NC: 357 2
[204] RCoNetks Cvd.Acc (avg.):97.89, Cvd.Sen(avg.):97.76, Cvd.Spe(avg.):98.24, Cvd.PPV(avg.):97.93, F1-score(avg.):97.63 N: 8851, C-19: 238, P: 6045 3

Cvd.Acc (%): COVID accuracy, Cvd.Sen(%): COVID sensitivity, Cvd.Spe(%): COVID specificity, Cvd.Pre(%): COVID precision, Normal: N, COVID-19: C-19, Pneumonia: P, Bacterial pneumonia: BP, Viral pneumonia: VP, Tuberculosis: TB, Non-COVID: NC, Others: O, Abnormal: AB, Private: Pvt., DS: dataset, Severe: S, Non-severe: NS, Mild: M, Moderate: mod, Critical: cr, Infected/Infection: I, Not infected: NI, Community acquired pneumonia (CAP): P, Lung cancer: LC.