Arabian Journal for Science and Engineering

https://doi.org/10.1007/s13369-023-07843-4

RESEARCH ARTICLE-COMPUTER ENGINEERING AND COMPUTER SCIENCE

Detection of Lung Opacity and Treatment Planning with

Three-Channel Fusion CNN Model
Fuat Türk1 · Yunus Kökver2

Received: 1 October 2022 / Accepted: 20 March 2023

© King Fahd University of Petroleum & Minerals 2023

Abstract
Lung opacities are extremely important for physicians to monitor and can have irreversible consequences for patients if
misdiagnosed or confused with other findings. Therefore, long-term monitoring of the regions of lung opacity is recommended
by physicians. Tracking the regional dimensions of images and classifying differences from other lung cases can provide
significant ease to physicians. Deep learning methods can be easily used for the detection, classification, and segmentation
of lung opacity. In this study, a three-channel fusion CNN model is applied to effectively detect lung opacity on a balanced
dataset compiled from public datasets. The MobileNetV2 architecture is used in the first channel, the InceptionV3 model in
the second channel, and the VGG19 architecture in the third channel. The ResNet architecture is used for feature transfer
from the previous layer to the current layer. In addition to being easy to implement, the proposed approach can also provide
significant cost and time advantages to physicians. Our accuracy values for two, three, four, and five classes on the newly
compiled dataset for lung opacity classifications are found to be 92.52%, 92.44%, 87.12%, and 91.71%, respectively.

Keywords Lung opacity detection · Deep learning · CNN · Three-channel fusion CNN model

1 Introduction dark-appearing lung that appear denser, hazy, or cloudy [2].

Therefore, areas of opacity are areas that are gray but should
Opacity is a term used to describe any area that preferen- be darker [3, 4].
tially absorbs and therefore appears more opaque than the Figure 1 shows a healthy lung image and an image with
surrounding area on a radiograph. It does not indicate the lung opacity.
size or pathological structure of the abnormality [1]. In other In lung opacity images, haziness can be seen in the areas
words, opacity refers to any area that appears white on a where labeled boxes are present (referred to as ground-
chest radiograph when it should be darker. On both CT and glass opacity) and the usual boundaries of the lungs are lost
chest radiographs, normal lungs appear dark compared to sur- (referred to as consolidation). Lung opacities are not homo-
rounding tissues because air has a relatively lower density. geneous and do not have a clear center or clear boundaries
When there is a replacement of air with another substance [6]. For this reason, it is difficult to separate it from the entire
in the lung, such as fluid or fibrosis, it causes an increase image and segment it properly [5].
in the density of that area. As a result, the tissue appears Lung opacity is generally benign and resolves sponta-
lighter or gray in a chest radiograph. The term “lung opac- neously without complications in patients with short-term
ity” on a chest radiograph refers to areas in the normally illness [7, 8]. The presence of opacity on a chest X-ray
image can indicate: fluid in air spaces, thickening of air space
B Yunus Kökver walls, thickening of lung tissue, inflammation, pulmonary
ykokver@ankara.edu.tr edema, damage and bleeding in blood vessels, cancerous
Fuat Türk growth, fibrosis [5, 9–11]. The increase in the area of opac-
fuatturk@karatekin.edu.tr ity also increases the risk of fatal pneumonia. The aim of
this research is to detect opacity on X-ray images, reduce
1 Department of Computer Engineering, Çankırı Karatekin the burden on hospitals and healthcare professionals, dis-
University, 18100 Çankırı, Turkey
tinguish between COVID-19, pneumonia, and tuberculosis,
2 Department of Computer Technologies, Elmadağ Vocational
School, Ankara University, 06780 Ankara, Turkey

123
 Arabian Journal for Science and Engineering

Fig. 1 X-ray images with

healthy and lung opacity [5] a
healthy image and b lung
opacity image

Fig. 2 Flowchart for the diagnosis of Lung Opacity

and encourage physicians to pay more attention to these areas tioned. The biggest difference of the dataset from other
before the disease reaches the pneumonia stage [12]. datasets is that it has a higher number of data in the lung
The main contributions of this paper are as follows: opacity, pneumonia, COVID-19, and normal categories.
Also, the maximum number of images possible have been
categorized for the tuberculosis class.
• Deep learning models have been successfully tested on • The three-channel fusion CNN model has been used for
X-ray images. An artificial intelligence system that helps the first time in the classification of diseases with lung
physicians diagnose lung opacity disease through lung images.
images has been developed. • The three-channel fusion CNN model has been proposed
• A dataset with five different classes has been created from as a new architecture that is easy to implement and has
datasets accepted in the literature. This dataset includes shown success in multi-class problems.
the lung opacity, pneumonia, COVID-19, tuberculosis,
and normal classes.
• The dataset is different from those in the literature in The remainder of this manuscript is organized as follows:
terms of the number and consistency of the data. It is a Section 2 introduces a literature review of lung opac-
unique dataset created for the five different classes men- ity diagnosis. Section 3 describes the materials and methods

123
Arabian Journal for Science and Engineering

Fig. 3 Multi-class model architecture

Table 1 Statistical information

Lung opacity Normal COVID-19 Pneumonia Tuberculosis
of the created dataset
Train 3306 7834 8107 7025 2088
Validation 1804 1679 1737 1505 448
Test 1804 1679 1738 1506 448
Total 6914 11,192 11,582 10,036 2984

Fig. 4 Samples from dataset

implemented in this study. Section 4 presents the experimen- rithms in a 1-dimensional vector of images, which produced
tal results and a comparison with the results in the literature. more representable features for individual diseases.
Section 5 discusses conclusions and future work. Li et al. [14] proposed the Cov-Net model for the detec-
tion of four-class (lung opacity, COVID-19, viral pneumonia,
and normal) radiological images. A modified residual net-
work with asymmetric convolution and embedded attention
mechanism was used as a backbone of the feature extractor
2 Literature Review for accurate detection of classes.
Mergen et al. [15] used deep learning methods for detect-
Sirazitdinov et al. [12] proposed an ensemble model consist- ing lung abnormalities. First, multi-scale deep reinforcement
ing of two convolutional neural networks, Mask R-CNN and learning was used for detecting anatomical landmarks. A
RetinaNet, for the localization and detection of lung opac- DenseUNet was trained for lung opacity segmentation.
ity and other pneumonia. Because of the study on 26684 Rahman et al. [16] used five image enhancement tech-
datasets, a recall value of 0.793 was obtained. niques to increase the accuracy of disease diagnosis on a
Senan et al. [13] used two deep learning models, AlexNet three-class (lung opacity, COVID-19, and healthy) 18479
and ResNet-50, to diagnose X-ray datasets created from mul- chest X-ray dataset. They then proposed a new UNet model
tiple sources. Enhanced features extracted from CNN models for lung segmentation. Six different pre-trained CNNs and
were then combined with traditional GLCM and LBP algo-

123
 Arabian Journal for Science and Engineering

Fig. 5 Lung opacity mask

extraction samples

Table 2 Features used for classification tasks a shallow CNN model were examined on both normal and
Model Three-channel fusion CNN segmented images.
Muhammad et al. [17] successfully applied deep learning
Image dimensions 224 × 224 with CNNs to a five-class (lung opacity, bacterial pneumonia,
Training options Adam viral pneumonia, COVID-19, and normal) dataset to increase
Max epochs 100 diagnostic accuracy. To augment data, due to the lack of X-ray
Activation function LeakyReLU& Softmax images, they proposed a self-augmentation mechanism using
Initial learning rate 0.0001 reconstruction independent component analysis (RICA).
Filter size 5×5
Maximum pooling 2×2
Batch size 32 3 Materials and Methods
Execution environment GPU
The motivation behind the diagnosis and segmentation study
Table 3 Average accuracy of lung opacity disease on chest X-rays is to help physi-
Class Accuracy (%)
values for different classes cians identify and follow the progression of the disease by
Five-class 91.71 using a deep learning technique-developed system to detect
Four-class 87.12 the lung opacity condition. The three-channel fusion CNN
Three-class 92.44 model has been used as a deep learning modeling algo-
Two-class 92.52 rithm to extract the most important distinguishing features
from the X-ray images. The images included in the lung
opacity class have also been segmented using python’s preex-

123
Arabian Journal for Science and Engineering

Fig. 6 Accuracy, loss, and

learning rate values for the
five-class classification process

Fig. 7 Accuracy, loss, and

learning rate values for the
four-class classification process

123
 Arabian Journal for Science and Engineering

Fig. 8 Accuracy, loss, and

learning rate values for the
three-class classification process

Fig. 9 Accuracy, loss, and

learning rate values for the
two-class classification process

123
Arabian Journal for Science and Engineering

Fig. 10 Confusion matrix for

the five-class dataset (0:
COVID-19, 1: lung opacity, 2:
normal, 3: pneumonia, 4:
tuberculosis)

isting libraries (OpenCV, matplotlib) to mask the image. The architectures available in TensorFlow have been tried in order
obtained mask images have been stored on a web server, and and the ones with the best results have been selected.) The
it has been planned for the physicians to interpret the differ- MobileNetV2 architecture has been used in the first chan-
ence between the new X-ray image of the patient taken after a nel, the InceptionV3 architecture in the second channel, and
certain period (between 1 year and 3 years) has passed. The the VGG19 architecture in the third channel. MobileNetV2
training results will be incorporated into the system using CNN is widely used in image classification and segmenta-
transfer learning to segment the lung opacity class in the tion processes [21, 22]. MobileNetV2 architecture works by
model. The flowchart of the study is shown in Fig. 2. reducing the size and complexity of the network in terms
of the number of parameters. For this reason, it is devel-
oped and preferred for efficiency. InceptionV3 is a modified
3.1 Model Description version of the inception family with some improvements
including LabelSmoothing, 7x7 convolutions as well. It is
The proposed CNN model is designed as a three-channel mostly used in image analysis and object detection problems
model. Classic fusion architecture models are used with two, [23]. VGG19 architecture is a deep neural network with mul-
three, and four channels [18, 19]. The basic idea of the tiple layers of convolution. It is useful due to its simplicity,
fusion architecture is to provide the input image multiple as it is composed of 3x3 convolutional layers stacked on top
times in multiple stages in order to extract more features of each other with increasing depth levels. To reduce the vol-
[20]. However, repeatedly providing the same image on dif- ume size, maximum pooling layers are used [24]. In each of
ferent channels can cause inconsistency in extracting more the three channels, the transfer of features from the previous
features. Therefore, the three channels in our proposed model layer to the current layer has been supported using ResNet
have been implemented with classical CNN models that have architecture.
been successful in classification problems. (In this stage, the

123
 Arabian Journal for Science and Engineering

Fig. 11 Confusion matrix for

the four-class dataset (0: lung
opacity, 1: normal, 2:
pneumonia, 3: tuberculosis)

During the transfer to the fully connected layer, the ated by Cohen et al. were also added to the pool of data
features from the three channels are combined with a con- [26]. The publicly available dataset created by Chowdhury
catenated layer and transferred to the output layer. Then, lung et al. was also examined and added to the data pool [16, 27].
opacity is detected with a multi-class classifier using the soft- Finally, the dataset created by Tawsifur et al. was added to
max function. Finally, the mask extraction process from the the data pool [28].
images belonging to the lung opacity class is performed with These are the final classes in the five-class categorization:
the help of the OpenCV and matplotlib libraries and saved lung opacity, pneumonia, COVID-19, tuberculosis, and nor-
on web servers. The architecture of the proposed model is mal (healthy). The number of images in each class included in
shown in Fig. 3. the dataset is shown in Table 1. The created dataset has been
made publicly available for the knowledge of researchers
[29]. More detailed descriptions of the dataset and a link to the
3.2 Dataset dataset can be found at: https://github.com/turkfuat/covid19-
pneumonia-dataset. Sample images used in the study are
The compiled dataset is a comprehensive version of the shown in Fig. 4.
data commonly used in literature. The number of images
belonging to classes that are difficult to learn (such as
pneumonia and lung opacity) has been kept as high as 3.3 Image Preprocessing
possible. First, the publicly available dataset created by
Deb and Jha [25] was examined and categorized. Then, As the chest X-ray images are obtained from different
images from the COVID-19 Grand Challenge dataset were sources, each image has different sizes, different contrasts,
included in these categorized groups (https://cxr-covid19. and different light reflections. Hence, the imaging intensity
grand-challenge.org/Dataset/). Images from the dataset cre- of each image is different. In addition, due to the lack of

123
Arabian Journal for Science and Engineering

Fig. 12 Confusion matrix for

the three-class dataset (0: lung
opacity, 1: normal, 2:
tuberculosis)

a certain standard in X-ray imaging and other reasons such using TensorFlow libraries [32, 33]. Some examples of seg-
as patient movements, noise occurs on the images. In noisy mentation are shown in Fig. 5.
images, the disease diagnostic accuracy of algorithms can be
reduced [30]. For this reason, preprocessing algorithms are
applied to the images in the dataset [31]. For this purpose,
the OpenCV Library in Python is used. If the pixel intensity 3.5 Evaluation Metrics
is less than the specified threshold value, the pixel is set to 0
(black) to prevent it from participating in the computations. Lung opacity detection is a classification task; therefore, the
The average filter is applied to enhance the images. The con- most fundamental metric that can be selected is the confusion
trast of each image is increased to expand the density range. matrix. The confusion matrix technique evaluates the accu-
All images are resized to a standard size of 224x224 pixels racy and performance of the classification algorithm. If the
for deep learning models. images in the classes of the dataset do not show a balanced
distribution, measuring the classification accuracy alone may
not be sufficient and may give misleading results [13, 34]. In
3.4 Lung Opacity Mask Extraction Process this study, the performance metrics calculated for the dataset
used are defined as accuracy, recall, precision, and F1 score.
After a lung opacity class is detected, the images in this class Accuracy is a measure of how well the algorithm is able
are masked in order to be compared with images obtained to correctly predict the class of a given sample. It is calcu-
within a range of one to three years. These masks will be used lated by dividing the number of correctly classified samples
to compare with new images obtained when physicians are by the total number of predictions made. In other words, it
called for control. For this purpose, the images are first con- represents the proportion of the total number of predictions
verted to DICOM format and then the lungs are segmented that the classifier got right.

123
 Arabian Journal for Science and Engineering

Fig. 13 Confusion matrix for

the two-class dataset (0: lung
opacity, 1: normal)

Recall is a measure of the performance of a classification TN + TP

Accuracy = (1)
model that indicates the proportion of actual positive cases TP + FP + TN + FN
that were correctly predicted by the model. It is particularly TP
Recall = (2)
useful when the classes are imbalanced, as it gives a more TP + FN
complete picture of the model’s performance on the minority TP
Precision = (3)
class. TP + FP
Precision is a measure of the performance of a classifica- 2 × (Recall × Precision)
F1 Score = (4)
tion model that indicates the proportion of predicted positive Recall + Precision
cases that were actually positive. In a confusion matrix, pre-
cision is calculated by dividing the number of true positive
predictions made by the model by the total number of pre- 4 Results and Discussion
dicted positive cases.
It is important to note that precision and recall are often The proposed model is initiated with the image dimensions
trade-offs of each other: Increasing one may result in decreas- set to 224x224 for the input layer. The ReLU and Leaky
ing the other. As such, it is often useful to consider both ReLU activation functions are applied to each channel, and
precision and recall when evaluating the performance of a Adam Optimizer and Stochastic Gradient Descent methods
classification model. One way to do this is to use the F1 are used as the optimizer. Leaky ReLU and Adam Optimizer
score, which is the harmonic mean of precision and recall. are chosen because they provided the best results among
These metrics are shown in Eq. (1) as accuracy, Eq. (2) as these methods. The learning rate is also tested at 0.001,
recall, Eq. (3) as precision, and Eq. (4) as F1 score. 0.003, 0.0001, and 0.0003. The best results are obtained with
0.0001, so this value is used for training. The filter size is set
to 5 × 5 and the maximum pooling to 2 × 2. The channels

123
Arabian Journal for Science and Engineering

Accuracy (%)
Table 4 Recall/ precision/ F1 score values for classifications
Class Recall Precision F1 score Samples

91.71
82.50
93.72

87.12
92.44
76.32

92.52
89.6
83.8
Five-class classification
Lung opacity 0.83 0.73 0.78 902
Normal 0.95 0.88 0.91 1679
COVID-19 0.96 0.95 0.95 1738

ResNet-50+SAM+BiLSTM
Three-channel fusion CNN
Three-channel fusion CNN
Three-channel fusion CNN
Three-channel fusion CNN
RetinaNet + Mask R-CNN
Pneumonia 0.90 0.99 0.94 3011
Tuberculosis 0.94 0.95 0.94 448
Four-class classification
Lung opacity 0.63 0.69 0.66 902

ResNet-50
Technique
Normal 0.96 0.86 0.91 1679

Cov-Net
AlexNet
Pneumonia 0.88 0.92 0.90 3011
Tuberculosis 0.95 0.94 0.95 448
Three-class classification
Lung opacity 0.84 0.91 0.88 902

Lung opacity, viral pneumonia, bacterial pneumonia, normal, COVID-19

Normal 0.96 0.91 0.94 1679
Tuberculosis 0.94 0.99 0.96 448
Two-class classification
Lung opacity 0.86 0.92 0.89 902

Lung opacity, pneumonia, tuberculosis, normal, COVID-19

Lung opacity, pneumonia, sparabreak normal, COVID-19
Normal 0.96 0.93 0.94 1679

Lung opacity, normal, Pneumonia, Tuberculosis

Lung opacity, pneumonia, normal, COVID-19
Lung opacity, pneumonia, normal, COVID-19
are combined in the concatenated layer. The output layer is
designed to be five-class with a softmax activation function.
The model was run with different numbers of epochs (60- Lung opacity, no lung opacity, normal

Lung opacity, normal, Tuberculosis

80-100-120). The training was terminated at 100 epochs, as
it was seen that the training did not progress further at this
stage. Hence, the training is set to 100 epochs. The training
of the network is completed in this way.

Lung opacity, normal

The GeForce GTX 1050 Ti graphics card is used for all
processes. The features used for classification algorithms are
shown in Table 2.
The average accuracy rate for the five-class classifica-
Classes

tion (lung opacity, normal, COVID-19, pneumonia, and

tuberculosis) using the three-channel fusion CNN model
is calculated to be 91.71%. The results of the four-class
(lung opacity, normal, pneumonia, and tuberculosis), three-
Number of classes

class (lung opacity, normal, and tuberculosis), and two-class

(lung opacity and normal) classifications are calculated to
Table 5 Comparison of the proposed study

be 87.12%, 92.44%, and 92.52%, respectively. The accuracy

results of the classification with different classes are shown
in Table 3.
5
4
3
3
4
4
4
5

Figure 6 shows the accuracy, loss, and learning rate values

for the five-class classification process. Accuracy, loss, and
learning rate values are shown for four-class classification
Sirazitdinov et al. [12]

Muhammad et al. [17]

in Fig. 7, three-class classification in Fig. 8, and two-class

classification in Fig. 9.
Senan et al. [13]
Senan et al. [13]
Previous studies

Proposed model
Proposed model
Proposed model
Proposed model

Table 4 shows the recall, precision, and F1 score perfor-

Li et al. [14]

mance of the classifiers.

Figure 10 shows the confusion matrix for detecting lung
opacity using a five-class and three-channel fusion CNN

123
 Arabian Journal for Science and Engineering

model. The confusion matrix of the four-class, three-class, References

and two-class classifiers created using the same model is
demonstrated in Figs. 11, 12, and 13, respectively. 1. Goodman, L.R.: Felson’s Principles of Chest Roentgenology, a
Programmed Text. Elsevier Health Sciences, eBook (2014)
Table 5 compares the proposed study with similar studies
2. Herring, W.: Learning Radiology: Recognizing the Basics. Elsevier
conducted in the literature. Health Sciences(2019)
Since Sirazitdinov et al. [12] developed a model on object 3. Chavez, S.; Long, B.; Koyfman, A.; Liang, S.Y.: Coronavirus dis-
recognition and used a smaller dataset, the accuracy values ease (covid-19): A primer for emergency physicians. Am. J. Emerg.
Med. 44, 220–229 (2021)
were more limited. Senan et al.’s [13] AlexNet and Resnet50
4. Rigby, D.-M.; Hacking, L.: Interpreting the chest radiograph.
models are less complex and have simpler structures com- Anaesth. Intensive Care Med. 22(6), 354–358 (2021)
pared to our three-channel model. In addition, their dataset 5. Zahavi, G.: What are lung opacities? https://www.kaggle.com/
only includes images from the viral pneumonia class in the zahaviguy/what-are-lung-opacities (2018)
6. Munden, R.F.; Black, W.C.; Hartman, T.E.; MacMahon, H.; Ko,
pneumonia category and does not include any images in the
J.P.; Dyer, D.S.; Naidich, D.; Rossi, S.E.; McAdams, H.P.; Good-
tuberculosis class, which leads to a higher classification accu- man, E.M.; et al.: Managing incidental findings on thoracic CT:
racy. Li et al. [14] have identified a performance similar to lung findings. A white paper of the acr incidental findings commit-
that of the model we proposed. The success of the Cov-Net tee. J. Am. Coll. Radiol. 18(9), 1267–1279 (2021)
7. Dennie, C.; Bayanati, H.; Souza, C.A.; Peterson, R.; Shamji, F.M.:
model in classification is noteworthy. However, the inclusion Role of the thoracic radiologist in the evaluation and management
of the tuberculosis class in the four-class classification again of solid and subsolid lung nodules. Thorac. Cardiovasc. Surg. 31(3),
increases the success. Muhammad et al. [17] conducted a 283–292 (2021)
five-class classification in which they divided the pneumo- 8. Ping-Chung, T.; Hsu, H.-S.; et al.: Management of pulmonary
ground glass opacity: a review of current clinical practice guide-
nia class into two and included viral pneumonia and bacterial lines. Formos. J. Surg. 55(3), 75 (2022)
pneumonia in the classification. The classification accuracy 9. Arenas-Jiménez, J.; Plasencia-Martínez, J.; García-Garrigós, E.:
was lower compared to other multi-class classifications. Cuando la neumonía no es covid-19. Radiologia 63(2), 180–192
(2021)
10. Kim, W.; Lee, S.M.; Kim, J.I.; Ahn, Y.; Park, S.; Choe, J.; Seo, J.B.:
Utility of a deep learning algorithm for detection of reticular opacity
on chest radiography in patients with interstitial lung disease. Am.
J. Roentgenol. 218(4), 642–650 (2022)
11. Zhu, M.; Yang, Z.; Wang, M.; Zhao, W.; Zhu, Q.; Shi, W.; Yu, H.;
Liang, Z.; Chen, L.: A computerized tomography-based radiomic
5 Conclusion and Future Work model for assessing the invasiveness of lung adenocarcinoma man-
ifesting as ground-glass opacity nodules. Respir. Res. 23(1), 1–11
Lung opacity, pneumonia, COVID-19, and tuberculosis are (2022)
often confused with each other. There is a need for serious 12. Sirazitdinov, I.; Kholiavchenko, M.; Mustafaev, T.; Yixuan, Y.;
Kuleev, R.; Ibragimov, B.: Deep neural network ensemble for
classification systems in these types of medical conditions pneumonia localization from a large-scale chest x-ray database.
where differences are not prominent. Therefore, we have Comput. Electr. Eng. 78, 388–399 (2019)
implemented a new approach for the detection and multi- 13. Senan, E.M.; Alzahrani, A.; Alzahrani, M.Y.; Alsharif, N.; Ald-
classification of lung opacity with a three-channel fusion hyani, T.H.: Automated diagnosis of chest x-ray for early detection
of covid-19 disease. Comput. Math. Methods Med. 2021 (2021)
CNN model. In this model, CNN models, which are success- 14. Li, H.; Zeng, N.; Wu, P.; Clawson, K.: Cov-net: a computer-aided
ful in classification problems, are designed so that each CNN diagnosis method for recognizing covid-19 from chest x-ray images
model works on a separate channel. Individual evaluation of via machine vision. Expert Syst. Appl. 207, 118029 (2022)
the extracted features and fusion steps provides a significant 15. Mergen, V.; Kobe, A.; Blüthgen, C.; Euler, A.; Flohr, T.; Frauen-
felder, T.; Alkadhi, H.; Eberhard, M.: Deep learning for automatic
advantage in the classification phase. Additionally, contrary quantification of lung abnormalities in covid-19 patients: first expe-
to the studies conducted with limited and unbalanced datasets rience and correlation with clinical parameters. Eur. J. Radiol. Open
in the literature, a dataset with a higher number of classes can 7, 100272 (2020)
produce stable results for training and testing. Therefore, our 16. Rahman, T.; Khandakar, A.; Qiblawey, Y.; Tahir, A.; Kiranyaz,
S.; Kashem, S.B.A.; Islam, M.T.; Al-Maadeed, S.; Zughaier, S.M.;
study can be a guide not only for lung opacity but also for Khan, M.S.; et al.: Exploring the effect of image enhancement tech-
other medical studies. niques on covid-19 detection using chest x-ray images. Comput.
In future studies, we plan to perform lung segmentation in Biol. Med. 132, 104319 (2021)
the new-progressive dataset that we will create for the follow- 17. Muhammad, U.; Hoque, M.Z.; Oussalah, M.; Keskinarkaus, A.;
Seppänen, T.; Sarder, P.: Sam: Self-augmentation mechanism for
up of cases detected with classification. It is believed that the covid-19 detection using chest x-ray images. Knowl.-Based Syst.
results can be integrated with the current study to create a 241, 108207 (2022)
web-based lung opacity warning system. In this way, we are 18. Huang, B.; Yang, F.; Yin, M.; Mo, X.; Zhong, C.: A review of mul-
optimistic that we can reduce the workload for physicians and timodal medical image fusion techniques. Comput. Math. Methods
Med. 2020 (2020)
medical institutions and provide a more comfortable living
environment for patients.

123
Arabian Journal for Science and Engineering

19. Li, M.-A.; Han, J.-F.; Yang, J.-F.: Automatic feature extraction and 29. Fuat, T.; Kökver, Y.: Application with deep learning models for
fusion recognition of motor imagery EEG using multilevel multi- covid-19 diagnosis. Sakarya Univ. J. Comput. Inf. Sci. 5(2), 169–
scale CNN. Med. Biol. Eng. Comput. 59(10), 2037–2050 (2021) 180 (2022)
20. Türk, F.; Lüy, M.; Barışçı, N.: Kidney and renal tumor segmentation 30. Ünver, H.M.; Kökver, Y.; Duman, E.; Erdem, O.A.: Statistical edge
using a hybrid v-net-based model. Mathematics 8(10), 1772 (2020) detection and circular hough transform for optic disk localization.
21. Kadam, K.D.; Ahirrao, S.; Kotecha, K.: Efficient approach towards Appl. Sci. 9(2), 350 (2019)
detection and identification of copy move and image splicing forg- 31. Duman, E.; Erdem, O.A.: Anomaly detection in videos using
eries using mask r-cnn with mobilenet v1. Comput. Intell. Neurosci. optical flow and convolutional autoencoder. IEEE Access 7,
2022 (2022) 183914–183923 (2019)
22. Howard, A.G.; Zhu, M.; Chen, B.; Kalenichenko, D.; Wang, W.; 32. Hofmanninger, J.; Prayer, F.; Pan, J.; Röhrich, S.; Prosch, H.;
Weyand, T.; Andreetto, M.; Adam, H.: Mobilenets: Efficient con- Langs, G.: Automatic lung segmentation in routine imaging is pri-
volutional neural networks for mobile vision applications. (2017) marily a data diversity problem, not a methodology problem. Eur.
arXiv preprint arXiv:1704.04861 Radiol. Exp. 4(1), 1–13 (2020)
23. Shadin, N.S.; Sanjana, S.; Lisa, N.J.: Covid-19 diagnosis from 33. Turk, F.; LÜY, M.; BARIŞÇI, N.; Yalcinkaya, F.: Kidney tumor
chest x-ray images using convolutional neural network (CNN) segmentation using two-stage bottleneck block architecture. Intell.
and inceptionv3. In: 2021 International Conference on Informa- Autom. Soft Comput. 33(1) (2022)
tion Technology (ICIT), pp. 799–804 (2021). IEEE 34. Ali, K.; Shaikh, Z.A.; Khan, A.A.; Laghari, A.A.: Multiclass skin
24. Mateen, M.; Wen, J.; Song, S.; Huang, Z.: Fundus image classifi- cancer classification using efficientnets–a first step towards pre-
cation using VGG-19 architecture with PCA and SVD. Symmetry venting skin cancer. Neurosci. Inf. , 100034 (2021)
11(1), 1 (2018)
25. Deb, S.D.; Jha, R.K.: Covid-19 detection from chest x-ray images
using ensemble of CNN models. In: 2020 International Conference
Springer Nature or its licensor (e.g. a society or other partner) holds
on Power, Instrumentation, Control and Computing (PICC), pp. 1–
exclusive rights to this article under a publishing agreement with the
5 (2020). IEEE
author(s) or other rightsholder(s); author self-archiving of the accepted
26. Cohen, J.P.; Morrison, P.; Dao, L.; Roth, K.; Duong, T.Q.; Ghas-
manuscript version of this article is solely governed by the terms of such
semi, M.: Covid-19 image data collection: prospective predictions
publishing agreement and applicable law.
are the future (2020) arXiv preprint arXiv:2006.11988
27. Chowdhury, M.E.; Rahman, T.; Khandakar, A.; Mazhar, R.; Kadir,
M.A.; Mahbub, Z.B.; Islam, K.R.; Khan, M.S.; Iqbal, A.; Al Emadi,
N.; et al.: Can AI help in screening viral and covid-19 pneumonia?
IEEE Access 8, 132665–132676 (2020)
28. Rahman, T.; Khandakar, A.; Kadir, M.A.; Islam, K.R.; Islam, K.F.;
Mazhar, R.; Hamid, T.; Islam, M.T.; Kashem, S.; Mahbub, Z.B.;
et al.: Reliable tuberculosis detection using chest x-ray with deep
learning, segmentation and visualization. IEEE Access 8, 191586–
191601 (2020)

123