Table 3.
State-of-the-art AI techniques to detect COVID-19 using CT scans.
| Paper | Method Used: Preprocessing + Segmentation + Feature Extraction + Feature Selection + Classification or CNN + Classification | Result Obtained | Dataset (Most Are Public) |
No. of Classes | |
|---|---|---|---|---|---|
| [205] | Various deep models are analyzed (Best: ResNet101) | Cvd.Acc: 99.51, Cvd.Sen: 100, Cvd.Spe: 99.02, AUC: 0.994 | C-19: 108,NC: 86,Total: 1020 slice, (Pvt.) | 2 | |
| [206] | EfficientNet family based architecture | Cvd.Acc: 98.99, Cvd.Sen: 98.80, Cvd.PPV:99.20 | DS 1- NC: 1230, C-19: 1252 | 2 | |
| Cvd.Acc: 56.16, Cvd.Sen: 53.06, Cvd.PPV: 54.74 (Train DS 1 & Test DS2) | DS 2: NC: 463,C-19: 349 | ||||
| [207] | LinkNet + DenseNet + DT | Cvd.Acc(avg.): 94.4, Cvd.Pre(avg.): 96.7, Cvd.Rec(avg.): 95.2, F1-score(avg.): 96.0 | C-19:445,NC:233 | 2 | |
| [208] | novel conditional generative model, called CoSinGAN | Independent testing is done using 50 CT cases (for lung segmentation and infection learning) | |||
| [93] | Intensity normalization and segmentation + Q-deformed entropy + ANOVA+ LSTM | Cvd.Acc: 99.68 | N: 107,C-19: 118,P: 96 | 3 | |
| [209] | Modified Alexnet model | Cvd.Acc: 94.75, Cvd.Sen: 93.22, Cvd.Spe: 96.69, Cvd.PPV:97.27 | C-19:3482,NC:2751 (Pvt.) | 2 | |
| [210] | Ensemble various models using majority voting scheme | Cvd.Acc: 85.2, Cvd.Sen: 85.4, Cvd.Pre: 85.7,F-score: 0.852,AUC: 0.91 | C-19 + ve: 349,C-19 -ve: 397 | 2 | |
| [211] | ResNet50 | Cvd.Acc: 82.91, Cvd.Sen: 77.66, Cvd.Spe: 87.62 | C-19:345,NC:397 | 2 | |
| [99] | CNN model with MODE | Cvd.Acc: outperforms competitive models by 1.9789% | 2 | ||
| [212] | Ensemble is built using ResNet152V2, DenseNet201, and VGG16 | Cvd.Acc: 98.83, Cvd.Sen: 98.83, Cvd.Spe: 98.82,F-measure: 98.30,AUC: 98.28 | N:3038,C-19:2373,P: 2890 TB: 3193 |
4 | |
| [36] | eXplainable Deep Learning approach (xDNN) | F1-score: 97.31 | SARS-CoV-2: 1252 Non SARS-CoV-2: 1230 |
2 | |
| [35] | Multi-task and self-supervised learning | Cvd.Acc: 89, F1- score: 0.90, AUC: 0.98 | C-19:349,NC: 463 | 2 | |
| [213] | Semi-Inf-Net | Cvd.Sen: 0.725, Cvd.Spe: 0.960, Dice: 0.739 | 100 images from 19 patients (Pvt) | C-19 lung Seg. | |
| [214] | 3D CNN model | Cvd.Acc: 87.50, Cvd.Sen: 86.90, Cvd.Spe: 90.10,F1-score: 82,AUC: 94.40 | Train: 2186, Test: 2796 (Pvt.) | 2 (CAP,C-19) | |
| [215] | CNN model | Cvd.Acc (avg): 94.03, Cvd.Sen(avg.): 94.44, Cvd.Spe (avg.): 93.63 | N: 320, C-19: 320 (Pvt.) | 2 | |
| [92] | AlexNet + Guided WOA | Cvd.Acc: 87.50, AUC: 99.50 | C-19: 334, NC-19: 794 | 2 | |
| [216] | Multi-task multi-slice deep learning system | Cvd.Acc: 95.21 | N: 251,C-19: 245,H1N1: 105 CAP: 123 (Pvt.) |
4 | |
| [217] | LBP and statistical features + ReliefF and NCA + DNN | Cvd.Acc: 95.84 | N: 397,C-19: 349 | 2 | |
| [218] | Region growing + deep CNN model (ResNet101 as its backbone) | Cvd.Acc: 94.9 | Total: 1110 patients with 5 classes | 5 | |
| [219] | Radiomic features + mRMR + XGBoost | AUC: 0.95 ± 0.02 | Total: 152 Patients | ||
| [220] | Segmentation of infectious lung as ResNet50 backbone | ||||
| [221] | DTCT and GLCM + RF | Cvd.Acc (avg.): 72.2, Cvd.Sen(avg.): 77, Cvd.Spe(avg.): 68,AUROC (avg.): 0.8 | C-19: 291, P: 279 (Pvt.) | 2 | |
| [222] | ResGNet (Graphs are generated using ResNet101-C features) | Cvd.Acc (avg.): 96.62, Cvd.Sen(avg.): 97.33, Cvd.Spe(avg.): 95.91, Cvd.Pre(avg.): 96.21,F1-Score(avg.): 0.9665 | N:148,C-19: 148 (Pvt.) | 2 | |
| [223] | CNN model (DenseNet201) + ELM | Cvd.Acc: 98.36, Cvd.Sen: 98.28, Cvd.Spe: 98.44, Cvd.Pre: 98.22,F1-Score: 98.25, AUC: 98.36 |
C-19: 349,NC: 397 | 2 | |
| [224] | M 2 UNet (Multi-task multi-instance deep network) | Cvd.Acc (avg.): 98.5, Cvd.Sen(avg.): 95.2, Cvd.Pre(avg.): 97.5,F1-Score(avg.): 0.963 AUC(avg.): 0.991 |
S:51,NS: 191(Pvt.) | 2 | |
| [225] | Dual-branch combination network (using UNet + ResNet50) | Cvd.Acc: 96.74, Cvd.Sen: 97.91, Cvd.Spe: 96.00,AUC: 0.9864 | N: 75 scans, C-19: 48 scans (Pvt.) | 2 | |
| [226] | Majority voting scheme with ResNet50 | Cvd.Acc: 96, Cvd.Sen:100, Cvd.Spe: 96,AUC: 0.90 | Two public datasets are used | 2 | |
| [227] | HE + WF + AlexNet + SVM | Cvd.Acc: 96.69, Cvd.Sen: 96, Cvd.Spe: 98 | N:500,C-19:488, P:500 | 3 | |
| [228] | DenseNet-201 | Cvd.Acc: 97.8, Cvd.Sen: 98.1, Cvd.Spe: 97.3, Cvd.Pre: 98.4, F1-score: 98.25 | C-19: 1500, NC: 1500 | 2 | |
| [229] | CLAHE + VGG-19 model | Cvd.Acc: 95.75, Cvd.Sen: 97.13,F1- score: 95.75, ROC-AUC: 99.30 | C-19 +ve: 1252, C-19 -ve: 1230 | 2 | |
| [230] | VGG16 model and ensemble learning | Cvd.Acc: 93.57, Cvd.Sen: 94.21, Cvd.Spe: 93.93, Cvd.Pre: 89.4,F1-score: 91.74 | N: 243,C-19: 790,P: 384 | 3 | |
| [61] | Z-score normalization and KF+CNN + fuzzy c-means + LDN | Cvd.Pre: 96, Cvd.Sen: 97, F-score: 97 and volume overlap error (VOE) of 5.6 ± 1:2%. | |||
| [231] | Golden Key Tool + VGG model | Cvd.Acc: 100 | DS1- N: 55, C-19: 349 | 2 | |
| Cvd.Acc: 93.478, Cvd.Pre: 97.33, F1-score: 87.5 | DS2- N: 55, C-19: 349, NC: 20 | 3 | |||
| Cvd.Acc: 90.12, Cvd.Pre: 90.6 | DS3- C-19: 349, NC: 396 | 2 | |||
| [232] | PatchShuffle Stochastic Pooling Neural Network (PSSPNN) | F1-score(avg.): 95.79 | Total:521 | 4(N,C-19, P, TB) | |
| [233] | Clinical information and chest CT features + XGBoost | Cvd.Sen: 90.91, Cvd.Spec: 97.96, AUC: 0.924 | Total: 198 | 2 (M,S) | |
| [234] | 3D CU-Net | DSC: 0.960, 0.963, 0.771, Cvd.Sen: 0.969, 0.966, 0.837, Cvd.Spe: 0.998, 0.998, 0.998 | C-19: 70 for detecting C-19 infection | ||
| [235] | Tensor + COVID-19-Net (VGG16) + Transfer-Net (ResNet50) | Cvd.Acc: 94, Cvd.Sen: 96, Cvd.Spe: 92 | N: 700, C-19: 700 | 2 | |
| [236] | Ensemble model (using Resnet18, Densenet201, Mobilenetv2 and Shufflenet) | Cvd.Acc: 96.51, Cvd.Sen: 96.96, Cvd.Spe: 96.00,F1-Score: 0.97,AUC: 0.99 | C-19: 349,NC: 397 | 2 | |
| [237] | LungINFseg, model for segmentation | Cvd.Acc (avg.): 98.92, Cvd.Sen(avg.): 83.10, Cvd.Spe(avg.): 99.52, DSC(avg.):80.34 intersection over union (IoU) (avg.): 0.6877 |
20 labeled COVID-19 CT scans (1800 + annotated Slices) |
||
| [238] | Feature Pyramid Network(FPN) DenseNet201 for detection | Cvd.Sen: 98.3 (m), Cvd.Sen: 71.2(mod), Cvd.Sen: 77.8(s), Cvd.Sen: 100(cr) | 1110 subjects Severity classification | ||
| [239] | Volume of interest based DenseNet-201 | Cvd.Acc: 88.88, Cvd.Sen:89.77, Cvd.Spe: 94.73, F1-Score: 88.88 | C-19: -moderate risk:40 severe risk:40 extreme risk:40 |
3 | |
| [240] | Various deep network architectures are analyzed using publicly available two COVID-19 CT datasets | 2 | |||
| [241] | UNet | F1-Score, improvement of 5.394 ± 3.015%. | +ve:492. -ve: 447 | ||
| [242] | Stationary wavelets + CNN model (Best: ResNet18) | Cvd.Acc: 99.4, Cvd.Sen: 100, Cvd.Spe: 98.6,AUC: 0.9965 | C-19:349, NC:397 | 2 | |
| [243] | Gabor filter + convolution and pooling layers + RF | F1 score: 0.99 | C-19: 349,NC: 397 | 2 | |
| [244] | Stacked autoencoder detector model | Cvd.Acc(avg.):94.7, Cvd.Sen(avg.):94.1, Cvd.Pre(avg.):96.54, F1-score (avg.):94.8 | C-19: 275,NC: 195 | 2 | |
| [245] | DenseNet201 model + k-NN | Cvd.Acc, Cvd.Sen, Cvd.Pre, & F1-score:100 | C-19:2740,Suspected Cases: 2740 (Private) | 2 | |
| [246] | CNN model + MI and Relief-F and DA +SVM | Cvd.Acc: 98.39, Cvd.Sen: 97.78, Cvd.Pre: 98.21, F1-score: 0.98, AUC: 0.9952 | SARS-CoV-2: 1252 Non SARS-CoV-2: 1230 |
2 | |
| Cvd.Acc: 90.0, Cvd.Sen: 84.06, Cvd.Pre: 93.55,F1-score: 0.8855, AUC: 0.9414 | C-19:349, NC: 463 | ||||
| [247] | VGG19 model | Cvd.Acc: 94.52 | C-19: 349,NC: 463 | 2 | |
| [248] | VGG16 model | Cvd.Acc: 98.0, Cvd.Sen: 99.0, Cvd.Spe: 94.9 | N: 275, C-19: 195 | 2 | |
| [249] | Radiological features + Chi-square test + Ensemble classifier | Cvd.Acc: 91.94, Cvd.Sen: 93.54, Cvd.Spe: 90.32,AUC: 0.965 | C-19: 306,non-COVID-19 pneumonia: 306 (Pvt.) | 2 | |
| [250] | Various CNN and texture based approaches | Cvd.Acc (avg.): 95.99, Cvd.Sen(avg.): 94.04, Cvd.Spe(avg.): 99.01,F1-score(avg.): 0.9284, AUC (avg.): 0.9903 | COVID-19: 386, NC: 1010 | 2 | |
| [251] | Worried deep neural network + pre-trained models (InceptionV3, ResNet50, and VGG19) | Cvd.Acc: 99.04, Cvd.Prec: 98.68, Cvd.Rec: 99.11,F-score: 98.90 | Total: 2623 (Pvt.) | 2(I,NI) | |
| [252] | Density peak clustering approach | Structural similarity index (SSIM): 89 | Total images: 12 (Pvt.) | C-19 Seg. | |
| [253] | EfficientNet-b0 model | Cvd.Acc: 99.83, Cvd.Sen: 92.86, Cvd.Spe: 98.32, Cvd.PPV:91.92 | Total images: 107,675 (Pvt.) | 2(C-19,NC) | |
| Cvd.Acc: 97.32, Cvd.Sen: 99.71, Cvd.Spe: 95.98, Cvd.PPV: 93.26 | 2 (C-19,P) | ||||
| [254] | EfficientNetB3 | Cvd.Sen: 97.2, Cvd.Spe: 96.8,F1-score: 0.970, AUC: 0.997 | N:105,C-19:143,P:147 (Pvt.) | 3 | |
| Cvd.Sen: 92.4, Cvd.Spe: 98.3,F1-score: 0.953,AUC: 0.989 | N: 121,C-19: 119, P: 117(Pvt.) | 3 | |||
| Cvd.Sen: 93.9, Cvd.Spe: 83.1,AUC: 0.954 | C-19: 856,Non-P: 254 (Pvt.) | 2 | |||
| [255] | COVID Segnet | For COVID-19 segmentation: Dice Score: 0.726, Cvd.Sen.: 0.751, Cvd.Pre.: 0.726 | Train: 731 Test: 130 patients (Pvt.) | Lung and infected regions seg. | |
| For lung segmentation: Dice Score: 0.987, Cvd.Sen.: 0.986, Cvd.Pre.: 0.990 | |||||
| [256] | Anam-Net | Dice Score: 0.956, Cvd.Acc.: 98.5, Cvd.Sen.: 92.7, Cvd.Spe.: 99.8 | N:929, AB:880 | Anomalies seg. | |