Table 12.
Comparison of results of the present study with those of previous studies.
| Study | Imaging Modality | Data set | Method | Task | Results (%) |
|
|---|---|---|---|---|---|---|
| Accuracy | AUC | |||||
| Ozturk et al. [6] | X-ray | 125 COVID-19/500 no finding | DarkCovidNet model with CNN structure | COVID-19 vs. no finding |
98.08 | – |
| Ozturk et al. [6] | X-ray | 125 COVID-19/500 no finding/500 pneumonia | DarkCovidNet model with CNN structure | COVID-19 vs. no finding vs.pneumonia |
87.02 | – |
| Toğaçar et al. [7] | X-ray | 295 COVID-19/65 normal/98 pneumonia | Model performing feature extraction with deep learning, optimization with SMO, classification with SVM |
COVID-19 vs. normal vs.pneumonia | 99.27 | – |
| Ucar and Korkmaz [8] | X-ray | 79 COVID-19/1583 normal/4290 pneumonia | Offline data augmentation, Bayesian optimization during DeepSqueezeNet training |
COVID-19 vs. normal vs. pneumonia | 98.26 | – |
| Apostolopoulos and Mpesiana [9] | X-ray | 224 COVID-19/504 normal/700 bacterial pneumonia | Transfer learning with VGG19 | COVID-19 vs. others | 98.75 | – |
| Apostolopoulos and Mpesiana [9] | X-ray | 224 COVID-19/504 normal/700 bacterial pneumonia | Transfer learning with VGG19 | COVID-19 vs. normal vs. bacterial pneumonia | 93.48 | – |
| Apostolopoulos and Mpesiana [9] | X-ray | 224 COVID-19/504 normal/714 pneumonia (bacterial and viral) | Transfer learning with MobileNet-v2 | COVID-19 vs. others | 96.78 | – |
| Apostolopoulos and Mpesiana [9] | X-ray | 224 COVID-19/504 normal/714 pneumonia (bacterial and viral) | Transfer learning with MobileNet-v2 | COVID-19 vs. normal vs. pneumonia | 94.72 | – |
| Butt et al. [10] | CT | 184 COVID-19/145 normal/194 influenza-A viral pneumonia | Model segmented the candidate region with 3D-CNN and classified using Noisy-OR Bayesian function |
COVID-19 vs. others | – | 99.6 |
| Li et al. [11] | CT | 1296 COVID-19/1735 CAP/1325 nonpneumonia | Model used ResNet50 for classification | COVID-19 vs. others | – | 96 |
| Kang et al. [12] | CT | 1495 COVID-19/1027 pneumonia | Model used VNet to segment the lesion region, extracted radiomic and hand-crafted features, and used five different classifiers for classification |
COVID-19 vs. pneumonia | 95.5 | – |
| Afshar et al. [13] | X-ray | 123 COVID-19/1341 normal/3845 pneumonia (bacterial and viral) | Model with capsule network structure | COVID-19 vs. normal vs. pneumonia | 98.3 | 97 |
| Mahdy et al. [14] | X-ray | 25 COVID-19/15 normal | Multilevel thresholding and classification with SVM | COVID-19 vs. normal | 97.48 | – |
| Hemdan et al. [15] | X-ray | 25 COVID-19/25 normal | Model with different deep learning structures | COVID-19 vs. normal | 90 | 90 |
| This study | X-ray | 80 COVID-19/80 normal/80 bacterial pneumonia | Framework involved radiomics, normalization, feature ranking, and GM-CPSO–NN | COVID-19 vs. others | 99.17 | 99.06 |
| This study | X-ray | 80 COVID-19/160 normal/160 bacterial pneumonia | Framework involved radiomics, normalization, feature ranking, and GM-CPSO–NN | COVID-19 vs. others | 99.25 | 99.53 |