Skip to main content
. 2024 Apr 29;24(9):2830. doi: 10.3390/s24092830

Table 11.

Accuracy comparison between our proposed work and existing literature.

Paper Dataset Model Classification Types Accuracy Limitations
Sanida et al. [10] COVID-19 Radiography Database (21,165 images) Preprocessing: Applied
Augmentation: Yes
Feature Extraction: N/A
Model: Modified VGG19		 Multi-class classification—(fibrosis, opacity, tuberculosis, normal, viral pneumonia, and COVID-19 pneumonia) Accuracy: 98.88%
Feature num: More than 100

  1. The limited number of images.

  2. Lack of noise and overlay text removal from images.

  3. Absence of information on optimal features.

  4. Lack of comparison with more recent state-of-the-art methods
Abubakar et al. [11] CT image datasets (328 common pneumonia, 1972 COVID-19, and 1608 healthy images) Image Preprocessing: Applied
Augmentation: Yes
Feature Extraction: HOG and CNN
Model:KNN, SVM		 Multi-class classification (COVID-19-positive, healthy, and common pneumonia) Accuracy: VGG-16 + HOG feature achieved 99.4% overall accuracy with SVM
Feature num: More than 100

  1. The limited number of images

  2. Lack of noise and overlay text removal from images.

  3. Absence of information on optimal features.

  4. Lack of comparison with more recent state of-the-art methods.
Kufel et al. [12] NIH ChestX-ray14 (112,120 images) Image Preprocessing: N/A
Augmentation: Yes
Feature Extraction: EfficientNet
Model: Transfer learning techniques		 Multi-class (15 classes—No Finding, Atelectasis, Cardiomegaly, Effusion, Infiltration, Mass, Nodule, Pneumonia, Pneumothorax, Consolidation, Edema, Emphysema, Fibrosis, Pleural thickening, Hernia) Accuracy: 84.28%
Feature num: More than 100

  1. Lack of noise and overlay text removal from images.

  2. Absence of information on optimal features.

  3. Low accuracy in multi-class classification.
Li et al. [13] 1. ChestX-Ray 14 (112,120 images)
2. CheXpert (224,316 image)		 Image Preprocessing: N/A
Augmentation: Yes
Feature Extraction: Res2Net50
Model: MLRFNet		 Multi-class (7 classes—Atelectasis, Effusion, Infiltration, Mass, Nodule, Pneumonia, Pneumothorax) Accuracy:
1. 85.30%
2. 90.40%
Feature num: More than 100

  1. Lack of noise and overlay text removal from images.

  2. Absence of information on optimal features.

  3. Low accuracy in multi-class classification.
Farhan et al. [14] 1. COVID-19 Radiography Database (C19RD) (2905 images)
2. Chest X-ray Images for Pneumonia (CXIP) (5856 images)		 Image Preprocessing: Applied
Augmentation: No
Feature Extraction: Res2Net50
Model: HDLA-DNN classifier		 Binary classification—disease (such as non-COVID-19 pneumonia, COVID-19 pneumonia) and healthy Accuracy:
1. 98.35%
2. 98.99%
Feature num: More than 100

  1. Lack of noise and overlay text removal from images.

  2. Absence of information on optimal features.

  3. Low accuracy in multi-class classification.
Nahiduzzaman et al. [15] ChestX-Ray14 dataset (29,871 images) Image Preprocessing: Applied
Augmentation: No
Feature Extraction: ELM
Model: CNN-ELM		 Multi-class (17 classes—Atelectasis, Cardiomegaly, Effusion, Infiltration, Mass, Nodule, Pneumothorax, Consolidation, Edema, Emphysema, Bacterial pneumonia, Viral pneumonia, COVID-19, Pleural thickening, Fibrosis, Hernia, and Tuberculosis) Accuracy:
90.92% for 17 lung diseases
99.37% for COVID-19
99.98% for TB
Feature num: More than 100

  1. Lack of proper image preprocessing.

  2. Lack of a proper augmentation technique.

  3. Absence of information on optimal features.

  4. Low accuracy in multi-class classification.

  5. Lack of comparison with more recent state-of-the-art methods.
Jin et al. [16] ChestX-ray14 (112,120 images) Image Preprocessing: N/A
Augmentation: No
Feature Extraction: DenseNet121
Model: CM-DML-GZSL		 Binary classification (COVID-19 and Non-COVID-19) Accuracy: 80.0%
Feature num: More than 100

  1. Lack of noise and overlay text removal from images.

  2. Lack of a proper augmentation technique.

  3. Absence of information on optimal features.

  4. Low accuracy in multi-class classification and
Tang et al. [17] 1. CXR dataset (6939 images)
2. CT dataset (85,725 images)		 Image Preprocessing: N/A
Augmentation: No
Feature Extraction: DenseNet201
Model: NSCGCN		 Binary classification (Infection and Normal) Accuracy:
1. 97.09%
2. 99.22%
Feature num: More than 100

  1. Lack of noise and overlay text removal from images.

  2. Lack of a proper augmentation technique.

  3. Absence of information on optimal features.

  4. Limited class classifications.
Shamrat et al. [5] Multiple sources (Total of 85,105 images) Image Preprocessing: Applied
Augmentation: N/A
Feature Extraction: N/A
Model: LungNet22		 Multi-class classification (10 classes-Control, COVID-19, Effusion, Lung Opacity, Mass, Nodule, Pulmonary Fibrosis, Pneumonia, Pneumothorax, Tuberculosis) After augmentation (80,000 images) Accuracy: 98.89%
Feature num: N/A

  1. Absence of optimal features Extraction.

  2. Lack of comparison with more recent state-of-the-art methods.
Guail et al. [9] Chest X-ray dataset from Kaggle (5856 images) Image Preprocessing: Applied
Augmentation: Yes
Feature Extraction: CNN
Model: PNA-GCN		 Binary classification (Pneumonia and Normal) Accuracy: 97.79%
Feature num: More than 100

  1. The limited number of images.

  2. Lack of a proper augmentation technique

  3. Absence of information on optimal features.

  4. Lack of multi-class classification.
Ragab et al. [18] Chest X-ray dataset from Kaggle (6310 images) Image Preprocessing: Applied
Augmentation: No
Feature Extraction: CNN
Model: CapsNet		 Multi-class classification (Pneumonia, Normal, and COVID-19) Accuracy:
1. 86.6% for normal,
2. 94% for Pneumonia,
3. 89% for COVID-19
Feature num: More than 100

  1. The limited number of images.

  2. Lack of a proper augmentation technique

  3. Absence of information on optimal features.

  4. Lack of multi-class classification.
Liang et al. [19] COVID-19 (399 images) Normal (400 images) Image Preprocessing: N/A
Augmentation: No
Feature Extraction: 3D-CNN
Model: GCN		 Binary classification (COVID-19 and Normal) Accuracy: 98.5%
Feature num: More than 100

  1. The limited number of images.

  2. Lack of a proper augmentation technique

  3. Absence of information on optimal features.

  4. Lack of multi-class classification.
Javaheri et al. [20] Not publicly available (16,750 slices of CT scan images from 335 patients) Image Preprocessing: N/A
Augmentation: N/A
Feature Extraction: N/A
Model: CovidCTNet		 Binary classification (COVID-19, non-COVID-19) Multi-class classification (COVID-19, CAP, control lungs) Accuracy:
1. 93.33% (Binary classification)
2. 86.66% (multi-class classification)

  1. Lack of image preprocessing.

  2. Low accuracy in multiclass Classification.

  3. lack of augmentation technique.
Alshazly et al. [21] SARS-CoV-2 CT Scan dataset (2482 images) and COVID-19-CT dataset (746 images) Image Preprocessing: N/A
Augmentation: N/A
Feature Extraction: N/A
Model:Different Deep learning models such as ResNet101 and DenseNet201.		 Binary classification (COVID and non-COVID) Accuracy:
1. 99.4% (ResNet101)
2. 92.9% (DenseNet201)
Feature num: N/A

  1. The limited number of images

  2. Lack of image preprocessing.

  3. Lack of comparison with more recent state-of-the-art methods.
(Our proposed work) Multiple resources (71,096 images) Image Preprocessing: Resizing, Denoising, CLAHE, De-annotation, Filtering
Augmentation: Elastic deformation
Feature Extraction: DCNN (proposed)
Model: DeepChestGNN (proposed)		 Multi-class classification (10 classes-Normal, Effusion, Pulmonary Fibrosis, Lung Opacity, Mass, Nodule, COVID-19, Pneumonia, Pneumothorax, Tuberculosis) After augmentation (70,000 images) Accuracy: 99.74%
Feature num: 100

  1. The experimentation with real images is not present.

  2. Lack of pixel-level image preprocessing and segmentation using markers.