Fig 1.
Chest x-ray images: (A) normal; (B) COVID-19 positive; (C) viral pneumonia.
Fig 2.
The pipeline process.
Fig 3.
Variation in chest x-ray images distribution.
Table 1.
Hyper-parameters of the build CNN model and preferred weights in this study.
Fig 4.
Feature extraction of the input image is performed via the convolution, ReLU and pooling layers, before classification by the fully connected layer.
Table 2.
The layers and layer parameters of the VGG16 model.
Table 3.
Hyper-parameters of the VGG16 model and preferred weights in this study.
Table 4.
Hyper-parameters of the InceptionV3 model and preferred weights in this study.
Table 5.
Hyper-parameters of the Xception model and preferred weights in this study.
Fig 5.
Proposed pre-trained method for COVID-19 detection.
Table 6.
Description of classification task.
Fig 6.
(A) and (C) original chest x-ray COVID-19 and Pneumonia images; (B) and (D) contrast enhanced image using Icomb.
Fig 7.
Visual feature maps in first layer and deep layer.
Table 7.
Confusion matrix.
Table 8.
Average performance of the pre-trained CNN models.
Table 9.
Average performance of the build CNN models.
Fig 8.
Confusion matrices for scenario I obtained by (A) build CNN and (B) VGG16.
Fig 9.
Confusion matrices for scenario II obtained by (A) build CNN and (B) VGG16.
Fig 10.
Visualisation on chest x-ray of normal/COVID-19/pneumonia infected using Grad-CAM on the proposed model.
Fig 11.
Comparison of accuracy achieved in selected models after fine-tune: (A) scenario I and (B) scenario II.
Fig 12.
Comparison of accuracy achieved in new CNN: Scenario I and scenario II.
Fig 13.
ROC curve in scenario I for (A) build CNN and (B) VGG16.
Fig 14.
ROC curve in scenario II for (A) build CNN and (B) VGG16.
Table 10.
Accuracy obtained by existing models and models used in the study.