Page 112 - IJEEE-2022-Vol18-ISSUE-1

108 | Majid & Ali

approaches. The (COVID-19/Normal) groups were included 2796 out of 500, while the proposed simple CGAN take less
throughout all experimental trials. The generated image to time and generates 5589 images in only three hours. Saman
data augmentation shows in figure 10. Motamed et al. [ 35], This work employed two forms of
GAN, namely IAGAN and DCGAN, yet the classification
The approach of the deep detection model was evaluated accuracy was 80% campier than the proposed method is
on various performance metrics, including accuracy (ACC), 97%. Loey et al. [15], Using a deep convolutional generative
total parameters, and numbers of image generation. Also, adversarial network to produce synthetic data] employed a
discussed many current deep learning approaches current CGAN with CT images rather than CXR images. It achieved
deep learning approaches that have been developed a testing accuracy of 0.82%.
mainly for COVID-19 classification and compared their
performance. Detailed performance metrics for the proposed CGAN trained on the original dataset that involved two
model and its benchmarked methods are shown in Table V. classes and then generate images for each category
depending on (given label). Figure 8 depicts the CGAN
training process regarding the generator and discriminator's
loss scores; the generator tries to reduce the loss function as
much as possible, which may fool a discriminator. CGAN
generated 2814 COVID-19 images. The number of epochs
was decided depending on the resolution of the generated
images, the quality of the produced images increased over
500 epochs and increased gradually until could generate
high-resolution images. CGAN was trained for many
iterations.

Fig.10: Sample the generated high-quality resolution
covid19 images by the proposed CGAN.

TABLE V Fig.11: The output of the CGAN at different time intervals.
The experiments of the suggested model and
other methodologies are presented in this section. Accuracy is a measure of the total number of correct
predictions produced by the model, and it is defined in the
study Image GAN Mean following ways.
type Type accuracy

Accuracy = ( TN + TP ) (2)

Abdul Waheed et (TN + FP + TP + FN )
al[14]
X-rays ACGAN 0.95

Loey et al[15] CT CGAN 0.82 When dealing with deep learning classification
situations where the output may be divided into two or more
classes, the Confusion Matrix is used as a performance
evaluation.

Amal Al-Shargabi X-rays CGAN 99.7
et al[16]
0.92
Mohd Asyraf et X-rays DCGAN 0.80
al[17] 0.80
IAGAN 0.97
Saman Motamed et X-rays +
al[35]
DCGAN

proposed model X-rays CGAN

Table V describes the evaluation of the present detection Fig.12: Confusion matrix with Covid-19 diagnosis using
models, the suggested augmented deep model, on the other CNN, utilizing synthetic data and actual data as inputs.
hand, was developed using X-ray images of the chest and a
huge number of samples. Amal Al-Shargabi et al. [16],
although the classification accuracy is high at 99%, the
CGAN used takes a lot of time to train. It needs 16:05:55 to
generate images; the number of images generated is only