Ідентифікація типів захворювання легень за допомогою згорткової нейронної мережі й архітектури VGG-16
Pneumonia, tuberculosis, and Covid-19 are different lung diseases but have similar characteristics. One of the reasons for the worsening of disease in lung sufferers is a diagnosis that takes a long time. Another factor, the results of the X-ray photos look blurry and lack contracture, causing diffe...
Збережено в:
| Дата: | 2023 |
|---|---|
| Автори: | , , , |
| Формат: | Стаття |
| Мова: | Англійська |
| Опубліковано: |
The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"
2023
|
| Теми: | |
| Онлайн доступ: | https://journal.iasa.kpi.ua/article/view/271023 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | System research and information technologies |
| Завантажити файл: |  |
Репозитарії
System research and information technologies| _version_ | 1867334430638473216 |
|---|---|
| author | Bukhori, Saiful Verdy, Bangkit Yudho Negoro Windi Eka, Yulia Retnani Januar, Adi Putra |
| author_facet | Bukhori, Saiful Verdy, Bangkit Yudho Negoro Windi Eka, Yulia Retnani Januar, Adi Putra |
| author_institution_txt_mv | [
{
"author": "Saiful Bukhori",
"institution": "University of Jember, Jember"
},
{
"author": "Bangkit Yudho Negoro Verdy",
"institution": "University of Jember, Jember"
},
{
"author": "Yulia Retnani Windi Eka",
"institution": "University of Jember, Jember"
},
{
"author": "Adi Putra Januar",
"institution": "University of Jember, Jember"
}
] |
| author_sort | Bukhori, Saiful |
| baseUrl_str | http://journal.iasa.kpi.ua/oai |
| collection | OJS |
| datestamp_date | 2023-11-07T22:19:24Z |
| description | Pneumonia, tuberculosis, and Covid-19 are different lung diseases but have similar characteristics. One of the reasons for the worsening of disease in lung sufferers is a diagnosis that takes a long time. Another factor, the results of the X-ray photos look blurry and lack contracture, causing different diagnostic results of X-ray photos. This research classifies lung images into four categories: normal lungs, tuberculosis, pneumonia, and Covid-19 using the Convolutional Neural Network method and VGG-16 architecture. The results of the research with models and scenarios without pre-trained use data with a ratio of 9:1 at epoch 50, an accuracy of 94%, while the lowest results are in scenarios using data with a ratio of 8:2 at epoch 50, non-pre-trained models, accuracy by 87%. |
| doi_str_mv | 10.20535/SRIT.2308-8893.2023.3.07 |
| first_indexed | 2025-07-17T10:28:03Z |
| format | Article |
| fulltext |
S. Bukhori, B.Y.N. Verdy, Y.R. Windi Eka, A.P. Januar, 2023
96 ISSN 1681–6048 System Research & Information Technologies, 2023, № 3
UDC 62-50
DOI: 10.20535/SRIT.2308-8893.2023.3.07
IDENTIFICATION OF LUNG DISEASE TYPES USING
CONVOLUTIONAL NEURAL NETWORK
AND VGG-16 ARCHITECTURE
S. BUKHORI, B.Y.N. VERDY, Y.R. WINDI EKA, A.P. JANUAR
Abstract. Pneumonia, tuberculosis, and Covid-19 are different lung diseases but
have similar characteristics. One of the reasons for the worsening of disease in lung
sufferers is a diagnosis that takes a long time. Another factor, the results of the
X-ray photos look blurry and lack contracture, causing different diagnostic results of
X-ray photos. This research classifies lung images into four categories: normal
lungs, tuberculosis, pneumonia, and Covid-19 using the Convolutional Neural Net-
work method and VGG-16 architecture. The results of the research with models and
scenarios without pre-trained use data with a ratio of 9:1 at epoch 50, an accuracy of
94%, while the lowest results are in scenarios using data with a ratio of 8:2 at epoch
50, non-pre-trained models, accuracy by 87%.
Keywords: tuberculosis, pneumonia, Covid-19, VGG-16, convolutional neural net-
work.
INTRODUCTION
Human internal organs that are often associated with external environmental fac-
tors, one of which is the lungs [1; 2]. The human lungs consist of two organs or a
pair, namely the right and left. The lungs are located in the thoracic region of the
human body, one of two large respiratory organs located in the chest cavity and
are responsible for adding oxygen and removing carbon dioxide from the blood.
The lungs have a rubbery texture and are pinkish-gray in colour. The lungs con-
sist of other tissues inside which function to exchange oxygen and carbon dioxide
[3]. Because of the process of exchanging oxygen and carbon dioxide, the lungs
are in contact with external environmental factors such as smoke, microbes, dust
and also chemicals in the environment as pollutants. The relationship with these
environmental factors increases the risk of lung disease [4].
Several diseases can attack the lungs. Common symptoms are shortness of
breath and coughing. Lung disorders can be acute or chronic. Several diseases
that can attack the lungs and related respiratory systems include bronchitis,
pneumonia, asthma, tuberculosis and Covid-19 [5]. Bronchitis is a respiratory dis-
ease that occurs as a result of an upper respiratory infection and is usually caused
by a virus [6]. Pneumonia is a respiratory disorder that causes inflammation of the
smallest parts of the lungs, namely the bronchioles and alveolar tissue. Asthma is
a disease caused by inflammation of the respiratory tract. This inflammation will
cause swelling and narrowing of the airways. Air that should flow into the lungs
becomes obstructed [7]. Tuberculosis is a bacterial infection caused by Mycobac-
terium tuberculosis which attacks and damages body tissues. Bacteria can be
transmitted through the airways. Tuberculosis generally attacks the lungs, but also
Identification of lung disease types using Convolutional Neural Network and VGG-16 …
Системні дослідження та інформаційні технології, 2023, № 3 97
has the risk of spreading to the lymph nodes, bones, central nervous system, heart
and other organs [8]. Covid-19 is an infectious disease caused by SARS-CoV-2,
a type of coronavirus [9]. Typical symptoms are fever, cough, flu and shortness of
breath. Covid-19 spreads from one person to another through droplets from the
respiratory tract which are often produced when coughing or sneezing. Droplet
range is usually up to 1 meter [10]. Droplets can stick to objects, but won't last
long in the air. The time from transmission of the virus to the onset of clinical
symptoms is between 1–14 days with an average of 5 days. Bronchitis, pneumo-
nia, asthma, tuberculosis and Covid-19 if not handled properly in a short time can
cause health complications [11].
Lung disease problems tend to increase due to delays in diagnosis. Diagnosis
takes a long time because of the similarities in the symptoms of lung disease. Ac-
cording to WHO, various lung diseases including pneumonia, tuberculosis and
Covid-19 have almost the same symptoms [12]. One of the main reasons for the
increase in lung disease problems during the Covid-19 pandemic is the long proc-
ess of diagnosis. Another factor is that X-Rays often appear blurry and have no
contractures, leading to a different diagnosis [13]. Additional laboratory test re-
sults are needed to identify whether it is classified as tuberculosis, pneumonia, or
Covid-19. One of the reasons for the unfavourable radiographic results is the dif-
ference in X-Ray intensity in photos of normal tissue and photos of glandular tis-
sue affected by lung disease [14]. To overcome this problem, image processing is
needed so that it can increase and improve image quality. A lung disease classifi-
cation system is needed to help diagnose lung disease quickly. Alternative
technologies that can be used to overcome this problem are the use of computer
vision and deep learning.
Computer vision is one part of artificial intelligence [15]. Computer Vision
is a technology that allows computers to recognize objects around them [16]. Sci-
ence and technology are developing very fast, especially in the development of
computer vision combined with deep learning. Deep learning can be used for de-
cision making, detecting diseases based on their symptoms and early detection
[17]. This research develops identification of lung disease types using Convolu-
tional Neural Network and VGG-16 architecture.
Several research have identified lung disease from chest X-Rays using small
volume datasets and applying machine learning [15; 18]. The results of applying
this technology are quite important for medical progress. Research related to the
early diagnosis and treatment of lung diseases using deep learning has also been
researched and the results are quite important in clinical treatment [19; 20; 21].
This research develops lung disease identification using Convolutional Neural
Network (CNN) and VGG-16 architecture. CNN with VGG-16 architecture has
higher accuracy than other network architectures in processing ImageNet datasets
[22]. In similar cases, classification of pneumonia from X-Ray images using
VGG-16 results 97.93 % accuracy, while classification of pneumonia from X-Ray
images using inception-V3 results 96.58% accuracy [23]. The inception-V3 is the
latest version developed from the inception-V1 and V2 models. The inception-V3
model is a CNN trained directly on a low-configuration computer. The training is
quite difficult, and takes much longer. This problem is solved through transfer
learning which saves the last layer of the model for the new category. The pa-
rameters of the previous layer are stored, and the inception-V3 model is decon-
structed when the last layer is removed using transfer learning techniques. CNNs
inception is a network in the form of a repeated convolution design configuration
S. Bukhori, B.Y.N. Verdy, Y.R. Windi Eka, A.P. Januar
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 98
pattern. The components in CNNs inception are input layer, 1×1 convolution
layer, 3×3 convolution layer, 5×5 convolution layer, max pooling layer, and con-
catenation layer [24].
Test results on the diagnosis of pneumonia showed that VGG-16 architecture
exceed Xception network at the accuracy with 87% and 82% respectively [25].
Xception is a development of inception. The inception model was developed with
depth wise separable convolution. The number of parameters is almost the same
as inception. Xception brings the inception hypothesis to eXtreme. First, the
cross-feature map is captured by a 1×1 convolution. Consequently, the correlation
of each channel is captured via a regular 3×3 or 5×5 convolution. This idea goes
to the extreme of doing 1×1 to each channel, then doing 3×3 to each output [26].
This is identical to replacing the inception module with depth wise separable
convolutions. With a higher level of accuracy, the VGG-16 architecture can
increase the value of lung disease classification based on CT-scan images. This
research also developed a classification of lung images into four classes, namely
normal lungs, tuberculosis, pneumonia, and Covid-19.
The rest of this paper is organized as follows: The proposed model for iden-
tification of lung disease types using CNN and VGG-16 architecture is discussed
in section 2. Section 3 provides a system design for identification of lung disease
types using CNN and VGG-16 architecture. Section 4 discusses the results and
analysis of CNN model and integrated system applications. Finally, conclusions
are given in section 5.
RESEARCH PROPOSED
CNN architecture developed using VGG-16. The proposed architecture has 13
convolution layers and 3 fully connected layers so that a total of 16 layers are
used, there are 5 max-polling layers which are forwarded to several convolution
layers. Each layer has a different image size, on the first layer the image will be
resized to 224224 , then on the next layer the image size will be reduced to
112112 , 5656 , 2828 , 1414 as shown in Fig. 1.
Fig. 1 shows that the first layer is a fully connected layer which has 4096
neurons, the second layer has 4096 neurons, and the third layer has 4 neurons. In
Fig. 1. The Proposed Model
Identification of lung disease types using Convolutional Neural Network and VGG-16 …
Системні дослідження та інформаційні технології, 2023, № 3 99
the last layer there are 4 layers according to the number of classifications, namely
normal, pneumonia, tuberculosis, and Covid-19. There are several layers in the
CNN architecture (Fig. 1) which are used in the training and testing process. This
architecture is designed to produce models with better accuracy.
Input image is the process of entering an input file in the form of a chest
X-Ray image. Before the data is entered into the training process, to prevent er-
rors in classification, the dataset is adjusted specifically for biased data or trun-
cated data. Inappropriate data will be discarded and appropriate data will be in-
cluded in a dataset that is ready to be classified. This process includes 3 stages,
namely changing the size of the photo with a target of 224224 pixels, the photo
data is separated into two parts, namely training data and testing data with a com-
parison ratio of 7 : 3, 8 : 2, and 9 : 1. Then the photo is converted from RGB to
grayscale.
The convolutional layer includes filter, kernel, stride, Relu activation, and
pooling operations. The convolution layer used in this research uses a kernel with
a size of 33 pixels and the number of strides is 1, and uses a padding
configuration. There are thirteen convolution layers, the first convolution layer
uses 64 filters, the kernel is 33 pixels. The size of the chest X-Ray is 224224
pixels which then uses the ReLu activation function. The second convolution
layer is similar to the first convolution process but continues with a pooling
operation with a strides size of 22 and a pooling size of 22 . The pooling
operation will produce an image measuring 112112 pixels and a total of 64
feature maps. This feature map will be included into the third convolution
process. The third convlution layer has different parameters from the previous
layer. The filter used in the third layer is 128 with a data input size of 112112
pixels. With a total of 128 filters, the third convolution feature map obtained is
128. The fourth convolution layer is similar to the third convolution process, but
the pooling operation is continued with a strides size of 22 and a pooling size
of 22 . The pooling operation will produce an image measuring 56 × 56 pixels
and a total of 128 feature maps. This feature map will be included in the in the
fifth convolution process. The fifth convolution layer has different parameters
from the fourth layer. There are 256 filters used in this fifth layer with an input
data size of 56 × 56 pixels. With a total of 256 filters, the convolution of the five
feature maps obtained is 256. The sixth convolution layer is similar to the fifth
convolution layer without changing any parameters. The filters used in the sixth
layer are 256 with a data input size of 56 × 56 pixels. The seventh convolution
layer has similarities with the fifth and sixth convolution layers without changing
any parameters. The filters used in the seventh layer are 256 with a data input
size of 5656 pixels. In this layer, pooling operations are continued with a
strides size of 22 and a pooling size of 22 . The pooling operation will
produce an image measuring 2828 pixels and a total of 256 feature maps. The
eighth convolution layer has different parameters from the previous layer. The
filter used in this eighth layer is 512 with a data input size of 2828 pixels.
Because there are 512 filters, the eighth convolution produces 512 feature maps.
The ninth convolution layer has the same parameters as the eighth convolution
layer. The filter used in the ninth layer is 512 and the input data is 2828 pixels.
Filters with a total of 512 in the ninth convolution produce a feature map totaling
512. The tenth convolution layer has the same parameters as the eighth and ninth
convolution layers. The filter used in the tenth layer is 512 and the input data is
2828 pixels. This process is followed by a pooling operation with strides size
of 22 and a pooling size of 22 . The pooling operation will produce an
S. Bukhori, B.Y.N. Verdy, Y.R. Windi Eka, A.P. Januar
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 100
image size of 1414 pixels and a total of 512 feature maps. The eleventh
convolution layer has similarities with the tenth layer, but has a different input
data size. The filter used in the eleventh layer is 512 with a data input size of
1414 pixels. There are 512 filters and in the eleventh convolution, the resulting
feature maps are 512. The twelfth convolution layer has the same parameters as
the eleventh convolution layer. The filter used is 512 with a data input size of
1414 pixels. These 512 filters will produce a feature map of 512. The thirteenth
convolution layer has the same parameters as the eleventh and twelfth layers. The
filter used is 512 with input data of 1414 pixels. Then the process is continued
with a pooling operation with a size of 22 strides and a pooling size of 22 .
The pooling operation will produce an image measuring c pixels and a total of
512 feature maps.
The flatten layer is the layer that converts the multidimensional array output
in the feature extraction process into a one-dimensional matrix which is then
followed by the fully connected layer process. While the fully connected layer
will be used as many as three fully connected layers. The first layer is 4096, the
second layer is 4096, and the last layer is 4 according to the classification de-
signed.
SYSTEM DESIGN
The system is designed using the CNN architecture with 16 layers according to
the VGG-16 architectural concept. The architectural design uses 13 convolution
layers and 3 fully connected layers so that the total layers used are 16 layers. This
research uses 5 layers of max-polling, which adjusts to several convolution layers.
Each layer has a different image size. The first layer will resize the image to
224224 , then the next layer will reduce the size of the image to a configuration
of 112112 , 5656 , 2828 , 1414 , and 77 , as shown in Table 1.
After modeling, it is continued
with testing of the CNN model that
has been designed to analyze the
accuracy of the model by changing
the data scenario, epoch and using
pre-trained models. This process
aims to select a model that has the
highest accuracy so that it can be
used for classification. Before
comparing accuracy, the model will
conduct data training and data
testing with a ratio of 7 : 3, 8 : 2
and 9 : 1.
The test was carried out by
changing the parameters to get
the best image results for the
classification of normal lungs, lungs
with tuberculosis, lungs with
pneumonia, and lungs with Сovid-
19. This research was conducted
using three different epochs,
namely: epoch 20, epoch 50, and
T a b l e 1 . Configuration VGG-16
No. Layer Output Shape
1 Conv2D 224, 224, 64
2 Conv2D 224, 224, 64
3 MaxPooling2D 112, 112, 64
4 Conv2D 112, 112, 128
5 Conv2D 112, 112, 128
6 MaxPooling2D 56, 56, 128
7 Conv2D 56, 56, 256
8 Conv2D 56, 56, 256
9 Conv2D 56, 56, 256
10 MaxPooling2D 28, 28, 256
11 Conv2D 28, 28, 512
12 Conv2D 28, 28, 512
13 Conv2D 28, 28, 512
14 MaxPooling2D 14, 14, 512
15 Conv2D 14, 14, 512
16 Conv2D 14, 14, 512
17 Conv2D 14, 14, 512
18 MaxPooling2D 7, 7, 512
19 Flatten 25088
20 Dense 4096
21 Dense 4096
22 Dense 4
Identification of lung disease types using Convolutional Neural Network and VGG-16 …
Системні дослідження та інформаційні технології, 2023, № 3 101
epoch 100.While the data scenario uses three different data scenarios from each
epoch that are used, namely scenario data 7 : 3, 8: 2 and 9: 1, and uses a
comparison between using the pre- training and without using the pre-training
model. Then the model will be compared for its level of accuracy and will be se-
lected through the model with the highest accuracy.
RESULTS AND ANALYSIS
The model that has been designed is tested to get the best scenario. Scenarios are
made based on the number of epochs and the amount of data. The number of ep-
ochs tested were 20, 50, and 100, while the amount of data used was a ratio of 7 :
3, 8 : 2, and 9 : 1. Model testing was also carried out on models with pre-training
and models without training.
Each scenario has a different value when using a different number of epochs,
even though the model used is the same. Comparison of the performance of the
CNN model as a whole and the test results are shown in Table 2 – Table 4 for
the non-pretrained model test cases; Table 5 – Table 7 for the test cases with the
pretrained model.
T a b l e 2 . Performance comparison with data ratio of 7 : 3 (non-pretrained)
Epoch
Performance
20 50 100
Data Scenario 7 : 3
Accuracy 87.33% 89% 90%
Precision 89% 90% 90%
Recall 87% 89% 90%
F1 score 87% 89% 90%
Learning Curve overfitting overfitting overfitting
The best performance on scenario data with a ratio of 7 : 3 (non–pretrained)
is at epoch 100 with an accuracy value of 90%, a precision value of 90%, a recall
value of 90%, and a F1 score value of 90%. The results of the learning curves in
this scenario have a high variance as shown in Fig. 2. The occurrence of a high
variance indicates that there has been overfitting.
1
2
1 —
2 —
Fig. 2. Learning Curve Data Scenario 7 : 3 Epoch 100
S. Bukhori, B.Y.N. Verdy, Y.R. Windi Eka, A.P. Januar
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 102
T a b l e 3 . Performance comparison with data ratio of 8 : 2 (non-pretrained)
Epoch
Performance
20 50 100
Data Scenario 8 : 2
Accuracy 87% 90.25% 92%
Precision 88% 90% 92%
Recall 87% 90% 92%
F1 score 87% 90% 92%
Learning Curve overfitting overfitting overfitting
The best performance on scenario data with a ratio of 8 : 2 (non-pretrained)
is at epoch 100 with an accuracy value of 92%, a precision value of 92%, a recall
value of 92%, and a F1 score value of 92%. The results of the learning curves in
this scenario have a high variance as shown in Fig. 3. The occurrence of a high
variance indicates that there has been overfitting.
T a b l e 4 . Performance comparison with data ratio of 9 : 1 (non-pretrained)
Epoch
Performance
20 50 100
Data Scenario 9 : 1
Accuracy 89.50% 94% 92%
Precision 89% 94% 91%
Recall 89% 94% 93%
F1 score 90% 94% 92%
Learning Curve overfitting overfitting overfitting
The best performance on scenario data with a ratio of 9 : 1 (non-pretrained)
is at epoch 50 with an accuracy value of 94%, a precision value of 94%, a recall
value of 94%, and a F1 score value of 94%. The results of the learning curves in
this scenario have a high variance as shown in Fig. 4. The occurrence of a high
variance indicates that there has been overfitting.
1
2
1 —
2 —
Fig. 3. Learning Curve Data Scenario 8 : 2 Epoch 100
Identification of lung disease types using Convolutional Neural Network and VGG-16 …
Системні дослідження та інформаційні технології, 2023, № 3 103
T a b l e 5 . Performance comparison with data ratio of 7 : 3 (pretrained)
Epoch Performance
20 50 100
Data Scenario 7 : 3
Accuracy 91% 91.17% 91.33%
Precision 91% 92% 92%
Recall 91% 91% 91%
F1 score 91% 91% 91%
Learning Curve overfitting overfitting overfitting
The best performance on scenario data with a ratio of 7 : 3 (pretrained) is at
epoch 100 with an accuracy value of 91.33%, a precision value of 92%, a recall
value of 91%, and a F1 score value of 91%. The results of the learning curves in
this scenario have a high variance as shown in Fig. 5. The occurrence of a high
variance indicates that there has been overfitting.
T a b l e 6 . Performance comparison with data ratio of 8 : 2 (pretrained)
Epoch
Performance
20 50 100
Data Scenario 8 : 2
Accuracy 91% 93.75% 92.50%
Precision 92% 94% 92%
Recall 91% 94% 92%
F1 score 91% 94% 93%
Learning Curve overfitting overfitting overfitting
Fig. 4. Learning Curve Data Scenario 9 : 1 Epoch 50
1
2
1 —
2 —
1
2
1 —
2 —
Fig. 5. Learning Curve Data Scenario 7 : 3 Epoch 100
S. Bukhori, B.Y.N. Verdy, Y.R. Windi Eka, A.P. Januar
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 104
The best performance on scenario data with a ratio of 8 : 2 (pretrained) is at
epoch 50 with an accuracy value of 93.75%, a precision value of 94%, a recall
value of 94%, and a F1 score value of 94%. The results of the learning curves in
this scenario have a high variance as shown in Fig. 6. The occurrence of a high
variance indicates that there has been overfitting.
T a b l e 7 . Performance comparison with data ratio of 9 : 1 (pretrained)
Epoch
Performance
20 50 100
Data Scenario 9 : 1
Accuracy 94% 95% 94.50%
Precision 94% 96% 93%
Recall 94% 95% 94%
F1 score 94% 95% 94%
Learning Curve overfitting overfitting overfitting
The best performance on scenario data with a ratio of 9 : 1 (pretrained) is at
epoch 50 with an accuracy value of 95%, a precision value of 96%, a recall
value of 95%, and a F1 score value of 95%. The results of the learning curves in
this scenario have a high variance as shown in Fig. 7. The occurrence of a
high variance indicates that there has been overfitting.
1
2
1 —
2 —
Fig. 6. Learning Curve Data Scenario 8 : 2 Epoch 50
1
2
1 —
2 —
Fig. 7. Learning Curve Data Scenario 9 : 1 Epoch 50
Identification of lung disease types using Convolutional Neural Network and VGG-16 …
Системні дослідження та інформаційні технології, 2023, № 3 105
CONCLUSIONS
Convolutional Neural Network classification model is tested by training the mod-
el using training data and testing the accuracy of the model with testing data.
The better the model, the better the accuracy obtained. Testing results show that
the CNN model has the highest accuracy in the scenario of non-pretrained data
model 9 : 1 at epoch 50 which is 94%, while the lowest result is at 8 : 2 epoch 50
scenario testing of data without using a pre-trained model with an accuracy
of 87%.
The model from this scenario will be used to be implemented into applica-
tions because it has the best accuracy among models with other scenarios. The
pulmonary infectious disease classification system classifies pulmonary infectious
diseases based on chest X-Rays uploaded by doctors/health workers with good
accuracy. This system can classify into four classes: Normal, Covid-19, Pneumo-
nia, and Tuberculosis. Pretraining models, epochs, and data scenarios can impact
model performance. This research has conducted tests to determine the best
performance achieved by using the VGG-16 architecture. The best performance is
obtained in the 9 : 1 data scenario, epoch 50 on the non pre-trained model, with an
accuracy value of 94%, precision value of 94%, recall value of 94%, and F1-score
value of 94%, while the lowest result is in the 8 : 2 data scenario test epoch 50 on
the non-pretrained model with an accuracy value of 87%, precision value of 88%,
recall value of 87%, and F1-score value of 87%.
REFERENCES
1. K.F. Budden et al., “Emerging pathogenic links between microbiota and the gut-
lung axis,” Nat. Rev. Microbiol., vol. 15, no. 1, pp. 55–63, 2017. doi:
10.1038/nrmicro.2016.142.
2. P.K. Wang L., F.H. Green, and S.M. Smiley-Jewell, “乳鼠心肌提取 HHS Public
Access,” Susceptibility aging lung to Environ. Inj., vol. 176, no. 1, pp. 539–553,
2010. doi: 10.1055/s-0030-1265895.
3. D.J. Randall, “Transport and exchange of respiratory gases in the blood | Carbon Di-
oxide Transport and Excretion,” Encycl. Fish Physiol., vol. 2, pp. 909–915, Dec.
2011. doi: 10.1016/B978-0-12-374553-8.00027-7.
4. F. Yu, R. Xiao, X. Li, Z. Hu, L. Cai, and F. He, “Combined effects of lung disease
history, environmental exposures, and family history of lung cancer to susceptibility
of lung cancer in Chinese non-smokers,” Respir. Res., vol. 22, no. 1, pp. 1–10, 2021.
doi: 10.1186/s12931-021-01802-z.
5. N. Clementi et al., “Viral respiratory pathogens and lung injury,” Clin. Microbiol.
Rev., vol. 34, no. 3, pp. 1–45, 2021. doi: 10.1128/CMR.00103-20.
6. U. Grote, A. Fasse, T.T. Nguyen, and O. Erenstein, “Food Security and the Dynam-
ics of Wheat and Maize Value Chains in Africa and Asia,” Front. Sustain. Food
Syst., vol. 4, no. February, pp. 1–17, 2021. doi: 10.3389/fsufs.2020.617009.
7. L.J. Quinton, A.J. Walkey, and J.P. Mizgerd, “Integrative physiology of pneumo-
nia,” Physiol. Rev., vol. 98, no. 3, pp. 1417–1464, 2018. doi:
10.1152/PHYSREV.00032.2017.
8. R. Gopalaswamy, S. Shanmugam, R. Mondal, and S. Subbian, “Of tuberculosis and
non-tuberculous mycobacterial infections - A comparative analysis of epidemiology,
diagnosis and treatment,” J. Biomed. Sci., vol. 27, no. 1, pp. 1–17, 2020. doi:
10.1186/s12929-020-00667-6.
9. M.T. Adil et al., “SARS-CoV-2 and the pandemic of COVID-19,” Postgrad. Med.
J., vol. 97, no. 1144, pp. 110–116, 2021. doi: 10.1136/postgradmedj-2020-138386.
S. Bukhori, B.Y.N. Verdy, Y.R. Windi Eka, A.P. Januar
ISSN 1681–6048 System Research & Information Technologies, 2023, № 3 106
10. L. Morawska and J. Cao, “Airborne transmission of SARS-CoV-2: The world should
face the reality,” Environ. Int., vol. 139, no. April, p. 105730, 2020. doi:
10.1016/j.envint.2020.105730.
11. M. Wei, Y. Zhao, Z. Qian, B. Yang, J. Xi, and J. Wei, “Pneumonia caused by Myco-
bacterium tuberculosis,” Microbes Infect., no. January, 2020.
12. “Clinical management of severe acute respiratory infection (SARI) when COVID-19
disease is suspected,” World Health Organization, 2020. [Online]. Available:
https://www.ptonline.com/articles/how-to-get-better-mfi-results
13. R. Yasin and W. Gouda, “Chest X-ray findings monitoring COVID-19 disease
course and severity,” Egypt. J. Radiol. Nucl. Med., vol. 51, no. 1, 2020. doi:
10.1186/s43055-020-00296-x.
14. A.M. O’Hare et al., “Complexity and Challenges of the Clinical Diagnosis and Man-
agement of Long COVID,” JAMA Netw. Open, vol. 5, no. 11, p. e2240332, 2022.
doi: 10.1001/jamanetworkopen.2022.40332.
15. A.I. Khan and S. Al-Habsi, “Machine Learning in Computer Vision,” Procedia
Comput. Sci., vol. 167, no. 2019, pp. 1444–1451, 2020. doi:
10.1016/j.procs.2020.03.355.
16. S. Wael, A. Elshater, and S. Afifi, “Mapping User Experiences around Transit Stops
Using Computer Vision Technology: Action Priorities from Cairo,” Sustain., vol. 14,
no. 17, 2022. doi: 10.3390/su141711008.
17. S. Arooj, S.U. Rehman, A. Imran, A. Almuhaimeed, A.K. Alzahrani, and A. Alzah-
rani, “A Deep Convolutional Neural Network for the Early Detection of Heart
Disease,” Biomedicines, vol. 10, no. 11, p. 2796, 2022. doi:
10.3390/biomedicines10112796.
18. Y. Erdaw and E. Tachbele, “Machine learning model applied on chest X-ray images
enables automatic detection of COVID-19 cases with high accuracy,” Int. J. Gen.
Med., vol. 14, pp. 4923–4931, 2021, doi: 10.2147/IJGM.S325609.
19. T.N. Sree Ganesh, R. Satish, and R. Sridhar, “Learning effective embedding for
automated COVID-19 prediction from chest X-ray images,” Multimed. Syst., no.
0123456789, 2022. doi: 10.1007/s00530-022-01015-4.
20. S.M. Fati, E.M. Senan, and N. ElHakim, “Deep and Hybrid Learning Technique for
Early Detection of Tuberculosis Based on X-ray Images Using Feature Fusion,”
Appl. Sci., vol. 12, no. 14, 2022. doi: 10.3390/app12147092.
21. V. Acharya et al., “AI-Assisted Tuberculosis Detection and Classification from
Chest X-Rays Using a Deep Learning Normalization-Free Network Model,” Com-
put. Intell. Neurosci., vol. 2022, p. 2399428, 2022. doi: 10.1155/2022/2399428.
22. H. Handoko, F. Rozy, F.E. Gani, and A. Dharma, “Comparative Analysis of Convo-
lutional Neural Network Methods in Detecting Mask Wear,” Budapest International
Research and Critics Institute-Journal (BIRCI-Journal), 5(2), 2022.
23. M. Mujahid, F. Rustam, R. Álvarez, J. Luis Vidal Mazón, I. de la T. Díez, and I.
Ashraf, “Pneumonia Classification from X-ray Images with Inception-V3 and
Convolutional Neural Network,” Diagnostics, vol. 12, no. 5, pp. 1–16, 2022. doi:
10.3390/diagnostics12051280.
24. G. Natarajan and P. Dhanalakshmi, “Classification of pneumonia using pre-trained
convolutional networks on chest X-Ray images,” Int. J. Health Sci. (Qassim), vol. 6,
no. April, pp. 5378–5390, 2022. doi: 10.53730/ijhs.v6ns1.6097.
25. E. Ayan and H.M. Ünver, “Diagnosis of pneumonia from chest X-ray images using
deep learning,” 2019 Sci. Meet. Electr. Biomed. Eng. Comput. Sci. EBBT 2019, pp.
10–13, 2019. doi: 10.1109/EBBT.2019.8741582.
26. Y.-J. Fan et al., “Machine Learning: Using Xception, a Deep Convolutional Neural
Network Architecture, to Implement Pectus Excavatum Diagnostic Tool from
Frontal-View Chest X-rays,” Biomedicines, vol. 11, no. 3, p. 760, 2023. doi:
10.3390/biomedicines11030760.
Received 05.01.2023
Identification of lung disease types using Convolutional Neural Network and VGG-16 …
Системні дослідження та інформаційні технології, 2023, № 3 107
INFORMATION ON THE ARTICLE
Saiful Bukhori, ORCID: 0000-0002-2527-1080, University of Jember, Indonesia, e-
mail: saiful.ilkom@unej.ac.id
Verdy Bangkit Yudho Negoro, University of Jember, Indonesia, e-mail: verdybang-
kit@gmail.com
Windi Eka Yulia Retnani, ORCID: 0009-0001-7838-0205, University of Jember, Indo-
nesia, e-mail: windi.ilkom@unej.ac.id
Januar Adi Putra, University of Jember, Indonesia, e-mail: januaradi.putra@unej.ac.id
ІДЕНТИФІКАЦІЯ ТИПІВ ЗАХВОРЮВАННЯ ЛЕГЕНЬ ЗА ДОПОМОГОЮ
ЗГОРТКОВОЇ НЕЙРОННОЇ МЕРЕЖІ Й АРХІТЕКТУРИ VGG-16 / Сайфул
Бухорі, Верді Бангкіт Юдхо Негоро, Вінді Ека Юліа Ретнані, Януар Аді Путра
Анотація. Пневмонія, туберкульоз і Covid-19 – різні захворювання легенів, але
мають схожі характеристики. Однією з причин загострення захворювання ле-
гень є довготривала діагностика. Іншим фактором є те, що результати рентге-
нівських знімків виглядають розмитими і з відсутністю контрактури, що спри-
чиняє різні результати діагностики рентгенівських знімків. Це дослідження
класифікує зображення легенів на чотири категорії, а саме: нормальні легені,
туберкульоз, пневмонія та Covid-19 за допомогою методу згорткової нейрон-
ної мережі та архітектури VGG-16. Результати дослідження з моделями та
сценаріями без попередньої підготовки використовують дані зі
співвідношенням 9:1 в епосі 50, точністю 94%, тоді як найнижчі результати в
сценаріях з використанням даних зі співвідношенням 8:2 в епосі 50, моделі без
попередньої підготовки, точність 87%.
Ключові слова: туберкульоз, пневмонія, Сovid-19, VGG-16, згорткова ней-
ронна мережа.
|
| id | journaliasakpiua-article-271023 |
| institution | System research and information technologies |
| keywords_txt_mv | keywords |
| language | English |
| last_indexed | 2025-07-17T10:28:03Z |
| publishDate | 2023 |
| publisher | The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" |
| record_format | ojs |
| resource_txt_mv | journaliasakpiua/0b/107f397b8e8a8fdc1dd165dff2392a0b.pdf |
| spelling | journaliasakpiua-article-2710232023-11-07T22:19:24Z Identification of lung disease types using convolutional neural network and VGG-16 architecture Ідентифікація типів захворювання легень за допомогою згорткової нейронної мережі й архітектури VGG-16 Bukhori, Saiful Verdy, Bangkit Yudho Negoro Windi Eka, Yulia Retnani Januar, Adi Putra tuberculosis pneumonia Covid-19 VGG-16 convolutional neural network туберкульоз пневмонія Сovid-19 VGG-16 згорткова нейронна мережа Pneumonia, tuberculosis, and Covid-19 are different lung diseases but have similar characteristics. One of the reasons for the worsening of disease in lung sufferers is a diagnosis that takes a long time. Another factor, the results of the X-ray photos look blurry and lack contracture, causing different diagnostic results of X-ray photos. This research classifies lung images into four categories: normal lungs, tuberculosis, pneumonia, and Covid-19 using the Convolutional Neural Network method and VGG-16 architecture. The results of the research with models and scenarios without pre-trained use data with a ratio of 9:1 at epoch 50, an accuracy of 94%, while the lowest results are in scenarios using data with a ratio of 8:2 at epoch 50, non-pre-trained models, accuracy by 87%. Пневмонія, туберкульоз і Covid-19 – різні захворювання легенів, але мають схожі характеристики. Однією з причин загострення захворювання легень є довготривала діагностика. Іншим фактором є те, що результати рентгенівських знімків виглядають розмитими і з відсутністю контрактури, що спричиняє різні результати діагностики рентгенівських знімків. Це дослідження класифікує зображення легенів на чотири категорії, а саме: нормальні легені, туберкульоз, пневмонія та Covid-19 за допомогою методу згорткової нейронної мережі та архітектури VGG-16. Результати дослідження з моделями та сценаріями без попередньої підготовки використовують дані зі співвідношенням 9:1 в епосі 50, точністю 94%, тоді як найнижчі результати в сценаріях з використанням даних зі співвідношенням 8:2 в епосі 50, моделі без попередньої підготовки, точність 87%. The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" 2023-09-29 Article Article application/pdf https://journal.iasa.kpi.ua/article/view/271023 10.20535/SRIT.2308-8893.2023.3.07 System research and information technologies; No. 3 (2023); 96-107 Системные исследования и информационные технологии; № 3 (2023); 96-107 Системні дослідження та інформаційні технології; № 3 (2023); 96-107 2308-8893 1681-6048 en https://journal.iasa.kpi.ua/article/view/271023/283978 |
| spellingShingle | туберкульоз пневмонія Сovid-19 VGG-16 згорткова нейронна мережа Bukhori, Saiful Verdy, Bangkit Yudho Negoro Windi Eka, Yulia Retnani Januar, Adi Putra Ідентифікація типів захворювання легень за допомогою згорткової нейронної мережі й архітектури VGG-16 |
| title | Ідентифікація типів захворювання легень за допомогою згорткової нейронної мережі й архітектури VGG-16 |
| title_alt | Identification of lung disease types using convolutional neural network and VGG-16 architecture |
| title_full | Ідентифікація типів захворювання легень за допомогою згорткової нейронної мережі й архітектури VGG-16 |
| title_fullStr | Ідентифікація типів захворювання легень за допомогою згорткової нейронної мережі й архітектури VGG-16 |
| title_full_unstemmed | Ідентифікація типів захворювання легень за допомогою згорткової нейронної мережі й архітектури VGG-16 |
| title_short | Ідентифікація типів захворювання легень за допомогою згорткової нейронної мережі й архітектури VGG-16 |
| title_sort | ідентифікація типів захворювання легень за допомогою згорткової нейронної мережі й архітектури vgg-16 |
| topic | туберкульоз пневмонія Сovid-19 VGG-16 згорткова нейронна мережа |
| topic_facet | tuberculosis pneumonia Covid-19 VGG-16 convolutional neural network туберкульоз пневмонія Сovid-19 VGG-16 згорткова нейронна мережа |
| url | https://journal.iasa.kpi.ua/article/view/271023 |
| work_keys_str_mv | AT bukhorisaiful identificationoflungdiseasetypesusingconvolutionalneuralnetworkandvgg16architecture AT verdybangkityudhonegoro identificationoflungdiseasetypesusingconvolutionalneuralnetworkandvgg16architecture AT windiekayuliaretnani identificationoflungdiseasetypesusingconvolutionalneuralnetworkandvgg16architecture AT januaradiputra identificationoflungdiseasetypesusingconvolutionalneuralnetworkandvgg16architecture AT bukhorisaiful ídentifíkacíâtipívzahvorûvannâlegenʹzadopomogoûzgortkovoínejronnoímerežíjarhítekturivgg16 AT verdybangkityudhonegoro ídentifíkacíâtipívzahvorûvannâlegenʹzadopomogoûzgortkovoínejronnoímerežíjarhítekturivgg16 AT windiekayuliaretnani ídentifíkacíâtipívzahvorûvannâlegenʹzadopomogoûzgortkovoínejronnoímerežíjarhítekturivgg16 AT januaradiputra ídentifíkacíâtipívzahvorûvannâlegenʹzadopomogoûzgortkovoínejronnoímerežíjarhítekturivgg16 |