IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 14, No. 4, August 2025, pp. 3192~3200

ISSN: 2252-8938, DOI: 10.11591/ijai.v14.i4.pp3192-3200  3192

A deep learning-based framework for automatic detection of

COVID-19 using chest X-ray and CT-scan images

Sivanagireddy Kalli1, Bukka Narendra Kumar2, Saggurthi Jagadeesh1,

Kushagari Chandramouli Ravi Kumar2
1
Department of Electronics and Communication Engineering, Sridevi Women’s Engineering College, Hyderabad, India
2
Department of Computer Science and Engineering, Sridevi Women’s Engineering College, Hyderabad, India

Article Info ABSTRACT

Article history: COVID-19 has profoundly impacted global public health, underscoring the
need for rapid detection methods. Radiography and radiologic imaging,
Received Aug 7, 2024 especially chest X-rays, enable swift diagnosis of infected individuals. This
Revised Apr 14, 2025 study delves into leveraging machine learning to identify COVID-19 from
Accepted Jun 8, 2025 X-ray images. By gathering a dataset of 9,000 chest X-rays and CT scans
from public resources, meticulously vetted by board-licensed radiologists to
confirm COVID-19 presence, the research sets a robust foundation.
Keywords: However, further validation is essential expanding datasets to encompass
enough COVID-19 cases enhances convolutional neural network (CNN)
Convolutional neural network accuracy. Among various machine learning techniques, deep learning excels
COVID-19 in identifying distinct patterns on imaging characteristics discernible in chest
CT images radiographs of COVID-19 patients. Yet, extensive validation across diverse
Deep learning datasets and clinical trials is crucial to ensure the robustness and
DenseNet-121 generalizability of these models. The conversation extends into
complexities, including ethical considerations around patient privacy and
integrating intelligent tech into clinical workflows. Collaborating closely
with healthcare professionals ensures this technology complements the
established diagnostic approach. Despite the potential to detect COVID-19
using chest X-ray imaging findings, thorough research and validation,
alongside ethical deliberations, are vital before implementing it in the
healthcare field. The results show that the proposed model achieved
classification accuracy and F1 score of 96% and 98%, respectively, for the
X-ray images.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Sivanagireddy Kalli
Department of Electronics and Communication Engineering, Sridevi Women’s Engineering College
Hyderabad, Telangana, India
Email: sivanagireddykalli@gmail.com

1. INTRODUCTION
Lung diseases span a vast array of conditions, impacting millions globally. Broadly categorized,
they affect lung tissues, blood flow, and airways. Disorders such as asthma, chronic obstructive pulmonary
disease (COPD), and pneumonia primarily obstruct airways, impairing the body’s oxygen absorption and
carbon dioxide expulsion. Blood flow conditions like pulmonary embolism and pulmonary hypertension
impede oxygen transport due to arterial clotting in the lungs. Additionally, diseases such as sarcoidosis and
pulmonary fibrosis further complicate respiratory health. The COVID-19 pandemic, instigated by the SARS-
CoV-2 virus, has heightened global concerns about respiratory health. Common symptoms include dry
cough, fatigue, respiratory distress ranging from mild to severe, loss of taste or smell, and fever. The virus

Journal homepage: http://ijai.iaescore.com

Int J Artif Intell ISSN: 2252-8938  3193

spreads through respiratory droplets expelled when infected individuals sneeze, cough, or talk, posing
significant risks to vulnerable populations, including those with underlying health conditions like diabetes,
chronic respiratory diseases, cardiovascular disorders, and cancer. Older adults with multiple health issues
are particularly susceptible to severe illness from COVID-19.
Diagnostic strategies for COVID-19 encompass antigen tests to detect current infection and
antibody tests to ascertain past exposure. Widely used PCR tests detect viral RNA from nasopharyngeal
swabs with high accuracy. However, global containment efforts face challenges due to limited testing
capacities, uneven resource distribution, and the absence of universally available vaccines or specific
treatments. In diagnosing COVID-19, medical imaging technologies play a crucial role, especially in rapidly
assessing lung abnormalities. CT scans and X-rays capture detailed chest images, with CT scans offering
three-dimensional views that enhance visualization of internal organs and precise disease location
identification. In contrast, X-rays provide two-dimensional images suitable for examining dense tissues but
lack depth perception. Despite the benefits of CT scans, access to high-quality imaging equipment varies
globally. Researchers increasingly focus on using chest CT scans to diagnose COVID-19 due to their
superior imaging capabilities. However, accurate interpretation requires skilled radiologists to promptly and
accurately identify COVID-19-related abnormalities.
To streamline COVID-19 diagnosis and reduce human intervention, researchers develop automated
diagnostic models. One innovative approach involves a three-stage methodology incorporating wavelet-
enhanced data augmentation, disease detection, and anomaly localization. This approach compensates for
limited COVID-19 CT images by pre-processing them using stationary wavelets to enhance features and
mitigate over-fitting. Subsequently, each image undergoes transformations like shearing, rotation, and
translation for effective dataset augmentation. In the second stage, transfer learning techniques classify CT
scans into COVID and non-COVID categories. Models like ResNet50, with additional convolutional layers
for enhanced feature extraction, optimize classification accuracy. The best-performing model is selected
based on comparisons with benchmark transfer learning models, ensuring robust performance in
distinguishing COVID-19-related anomalies from normal lung images. In the final stage, the selected
model’s feature maps and activation layers detect anomalies within chest CT images of COVID-19-positive
patients, facilitating early and accurate clinical intervention. This research aims to evaluate and compare
transfer learning models' performance in detecting COVID-19 from CT scans using a limited dataset.
Additionally, the study explores innovative data augmentation techniques to enhance feature map
interpretability in deeper neural network layers, improving diagnostic accuracy and reliability. Advancements
in medical imaging and artificial intelligence hold promise for enhancing COVID-19 diagnosis and
management. By leveraging automated diagnostic models and innovative image processing techniques,
researchers aim to overcome existing challenges, expedite COVID-19 infection identification from chest CT
scans, support timely healthcare interventions, and mitigate the pandemic's impact on global health systems.

2. LITERATURE SURVEY
The COVID-19 pandemic, caused by the SARS-CoV-2 virus, has strained healthcare systems
worldwide, necessitating efficient and accurate diagnostic tools. Early and accurate diagnosis is crucial for
effective treatment and containment. While RT-PCR tests are the standard, imaging techniques like chest
X-rays and CT scans have proven invaluable in diagnosing and monitoring lung abnormalities associated
with COVID-19. Recent advancements in deep learning offer promising solutions for automating and
enhancing these diagnostic processes [1]. Chest X-rays provide a quick and accessible method for detecting
lung abnormalities, although they offer limited depth perception. CT scans, on the other hand, provide
detailed three-dimensional images, allowing for a more comprehensive assessment of lung conditions.
Studies have shown that CT scans are more sensitive in detecting COVID-19-related lung abnormalities
compared to chest X-rays [2]. COVID-19 exhibits distinct imaging features on chest X-rays and CT scans,
including ground-glass opacities, consolidation, and bilateral lung involvement. Identifying these features
accurately requires skilled radiologists, which can be a bottleneck in areas with limited medical expertise [3].
Convolutional neural networks (CNNs) have revolutionized image analysis by automating feature extraction
and classification processes. These networks have shown remarkable success in medical imaging tasks,
including disease detection and anomaly localization [4]. Transfer learning involves pre-training a neural
network on a large dataset and fine-tuning it on a specific task. This approach is particularly useful in
medical imaging, where labeled data is often scarce.
Models like ResNet, VGG, and densely connected convolutional networks (DenseNet) have been
effectively employed in transfer learning for COVID-19 detection [5]. Several studies have explored the use of
deep learning models for automatic COVID-19 detection from chest X-rays. Apostolopoulos and Mpesiana [6]
developed a CNN-based model achieving high accuracy in classifying COVID-19 cases. They used a dataset
of chest X-rays including COVID-19, pneumonia, and healthy cases, demonstrating the potential of CNNs in

A deep learning-based framework for automatic detection COVID-19 using chest … (Sivanagireddy Kalli)
3194  ISSN: 2252-8938

distinguishing between these conditions. Ozturk et al. [7] introduced a deep learning model capable of
detecting COVID-19 from chest X-rays with an accuracy of 98.08%. Their approach employed a DarkNet
model, highlighting the effectiveness of deep learning architectures in medical image analysis. CT scans offer
higher sensitivity and specificity in detecting COVID-19-related lung abnormalities. Wang et al. [8] proposed
a deep learning model using a dataset of chest CT images to detect COVID-19 with an accuracy of 82.9%.
Their model utilized a combination of CNN and long short-term memory (LSTM) networks to capture spatial
and temporal features from CT scans. Jaiswal et al. [9] employed a transfer learning approach using the
VGG16 model, achieving an accuracy of 91% in classifying COVID-19 from chest CT images. Their study
demonstrated the importance of data augmentation and preprocessing in enhancing model performance.
Comparative studies have shown that combining chest X-rays and CT scans can improve diagnostic accuracy.
For instance, Song et al. [10] developed a hybrid model integrating features from both imaging modalities,
achieving superior performance compared to models using a single modality. Their research emphasizes the
potential of multi-modal approaches in medical imaging. A study by Yan and Liming [11] evaluated chest
X-rays' performance in diagnosing COVID-19 and assessed radiologist interpretations' accuracy. The unique
CT characteristics of COVID-19 provided crucial insights for healthcare professionals on distinguishing
COVID-19 from other viral pneumonias based on imaging findings. The emergence of the novel coronavirus,
SARS-CoV-2, underscored the importance of accurate and timely diagnostic imaging in managing global
health crisis. Transfer learning has proven effective in addressing the anomaly detection challenge in small
medical image datasets, demonstrating promising results in distinguishing COVID-19 cases from other
respiratory conditions based on image features. One of main challenges in developing robust deep learning
models for COVID-19 detection is scarcity and quality of annotated datasets. Large-scale, diverse datasets
are crucial for training models that generalize well across different populations and imaging devices [12].
The well-known Deep CNN establishes the critical benchmarks by Wang et al. [8], and Min et al. [13]
proposed network in network, enhancing CNNs with micro multilayer perceptron (MLP) for improved
feature abstraction and classification accuracy. Nishiura et al. [14] analyzed COVID-19 serial intervals,
providing essential data for transmission modeling and early outbreak response strategies. Images directly
obtained from patients suffering with severe COVID-19 or pneumonia are used in this study [15]–[19]. The
lack of CT scans with the label "data" in radiology [20]. Additionally, the pretrained CNN model and texture
descriptors [21]. The goal of the Imaging COVID-19 AI initiative in Europe [22], [23] and the radiological
society of North America (RSNA) [24], [25] is to make data easily accessible to the general public. With the
help of these data, different features from different categories can improve interclass variance, which
improves deep learning performance. The model will overfit and yield conclusions that are only weakly
generalized due to a paucity of data [26], [27]. Therefore, it has been demonstrated that data augmentation
works well for training discriminative deep learning models. Flipping, rotating, color jittering, random
cropping, elastic distortions, and synthetic data synthesis using generative adversarial networks (GANs) are a
few examples of data augmentation techniques [28], [29]. Several visual traits of the medical photos in
ImageNet exhibit strong interclass similarity [30], [31]. As a result, conventional augmentation techniques
that just make minor image adjustments are less successful [32].

3. PROPOSED METHOD
In medical imaging, deep learning techniques have emerged as powerful tools for automating
diagnostics and improving disease detection. Automated COVID-19 diagnosis using medical imaging has
been explored using datasets comprising chest X-ray images from patients with bacterial pneumonia,
confirmed COVID-19 cases, and uninfected controls. Deep neural network architectures have been
investigated to enhance medical image classification accuracy.
DenseNet is a deep learning architecture that was developed to solve some of the shortcomings of
CNNs. These problems include vanishing gradients, information loss, and challenges in training extremely
deep network. The ability of the DenseNet design to effectively learn from data, eliminate difficulties with
vanishing gradients, and achieve great performance with relatively fewer parameters than other architectures
has contributed to the rise in popularity of this particular architecture. The number "121" in DenseNet121
refers to the total number of layers, which includes all fully connected, convolutional, pooling, and batch
normalization layers. DenseNet121, a CNN architecture, has several notable benefits in diverse computer
vision and image analysis applications. One notable benefit is in its extensive connection network, which
facilitates the reuse of features and the efficient transmission of information across different layers. In CNNs,
the process of combining feature maps is typically performed in a sequential manner. However, DenseNet121
introduces a novel approach where each layer establishes direct connections with all following layers. The
high level of connectedness inside the network facilitates the transmission of gradients, mitigates the
likelihood of disappearing gradients, and augments the network's total capacity for learning. Consequently,

Int J Artif Intell, Vol. 14, No. 4, August 2025: 3192-3200

Int J Artif Intell ISSN: 2252-8938  3195

DenseNet121 has a tendency to need a reduced number of parameters in comparison to other deep designs,
making it computationally effective and more amenable to training, especially when confronted with little
data. The compact nature of this feature makes it especially advantageous in situations when there are limited
processing resources available. One further benefit of DenseNet121 lies in its remarkable efficacy in the
processes of feature extraction and representation learning. The network's tight interconnections facilitate the
extraction of subtle and highly detailed information from pictures, rendering it particularly advantageous for
jobs demanding meticulous analysis, such as image categorization, object recognition, and medical image
analysis. Proposed deep learning DenseNet architecture as shown in Figure 1.
In addition, the inclusion of dense skip connections enables the establishment of skip connections
across various levels within the network, hence assisting in the retention of low-level characteristics and
contextual information over the whole of the network. This approach proves to be very beneficial in the
context of tasks such as semantic segmentation, where the availability of pixel-level information is of utmost
importance. In several computer vision applications, the popularity and success of DenseNet121 may be
attributed to its efficient parameter utilization and feature extraction capabilities. DenseNet121 is an example
of a CNN design that has been gaining traction in the world of medical imaging, namely in the identification
of COVID-19 utilizing chest X-ray pictures. This is one of the most important applications of this kind of
network. The significance of DenseNet121 in the detection of COVID-19 using chest X-ray images lies in
the fact that it is able to effectively learn and analyze features from medical images, provide state-of-the-art
performance, and ensure data efficiency and interpretability, all of which are essential for accurate and
trustworthy disease diagnosis.

Figure 1. Proposed deep learning DenseNet architecture

4. EXPERIMENTAL RESULTS
In this section, the results of the suggested method are analyzed and examined. Kaggle was the
source of the dataset that was used. The dataset was processed using the approach that was proposed. The
dataset is structured with three main folders called train, test, and val, and within those three main folders are
subfolders labeled "Pneumonia" and "Normal." There are 5,863 X-Ray pictures in the JPEG format, and they
are separated into two categories: pneumonia and normal. Chest X-ray pictures were chosen from
retrospective cohorts of pediatric patients aged one to five years old at the Guangzhou Women and Children's
Medical Center in Guangzhou. The photos were taken in an anterior-posterior orientation. In the course of
providing normal clinical treatment for the patients, all chest X-ray imaging was conducted. Before doing the
analysis of chest x-ray pictures, every chest radiograph was first subjected to a screening for quality control.
This included deleting any scans that were of poor quality or could not be read. After that, the diagnoses for
the photos were scored by two highly qualified doctors before being given the green light for use in the
A deep learning-based framework for automatic detection COVID-19 using chest … (Sivanagireddy Kalli)
3196  ISSN: 2252-8938

training of the AI system. In order to take into consideration, the possibility of grading mistakes, the
assessment set was also reviewed by a third specialist.
The ROC curve is created by comparing the true positive rate (TPR) to the false positive rate (FPR),
which is plotted against each other on the x- and y-axes, respectively. Every point on the ROC curve
represents a different classification threshold that is applied to the probabilities that are generated by the
model. We are able to regulate the balance between sensitivity and specificity by modifying the threshold,
and this may be done in accordance with the needs of the issue.
The Figure 2 is a graphical illustration of how well a binary classifier differentiates between the two
categories in question. A classifier that achieves high sensitivity (TPR) while simultaneously retaining a low
FPR is considered to have greater performance. This is shown by a curve that is closer to the top-left corner.
It is common practice to utilize the area under the ROC curve (AUC-ROC) as a summary indicator for the
overall performance of the model. A greater capacity for categorization is indicated by an AUC-ROC value
that is closer to 1.0. On the ROC graph, the diagonal line at 45 degrees depicts random guessing. This is the
case when the TPR and the FPR are identical to one another. A classifier that falls below this diagonal is
inferior to random guessing in terms of its predictive power. In contrast, a classifier that is located above the
diagonal has some degree of predictive power. The discriminating power of the model increases in proportion
to the steepness of the curve's ascent toward the upper left corner. The confusion matrix is a commonly used
tabular representation that is utilized to evaluate the efficacy of a classification model.
The assessment of the degree to which a model's predictions correspond with the observed results is
an important and useful technique, particularly in scenarios where there is an unequal distribution of classes.
In Figure 3, shows the matrix presented seems to pertain to a binary classification task, whereby the classes
are denoted as "Normal" and "COVID." The row in question pertains to situations in which the actual class is
classified as "Normal." In the present scenario, the model has accurately classified 265 cases as "Normal,"
but it has made erroneous predictions by classifying 4 instances as "COVID" when they were really
"Normal." The term "Actual COVID" refers to cases in which the true class is labeled as "COVID." The
model has accurately classified 67 cases as "COVID," but it has made erroneous predictions in 3 instances by
classifying them as "Normal" when they were in fact "COVID." The column labeled "Predicted normal"
denotes cases that have been classified as "Normal" by the model. Among the occurrences classed as
"Normal," a total of 265 were accurately labeled as such, while 3 instances were erroneously classified as
"Normal" when they were really cases of "COVID." The column labeled "Predicted COVID" denotes the
occurrences that the model has identified and classified as "COVID." Among the occurrences that were
categorized as "COVID," a total of 67 instances were accurately labeled as "COVID," whereas 4 instances
were erroneously classified as "COVID" when they were really examples of "Normal." The number of
instances accurately classified as "COVID" is 67, which are referred to as true positives (TP). True negatives
(TN) refer to instances that have been accurately forecasted as "Normal." In this specific case, there are 265
instances that have been properly classified as "Normal." False positives (FP) refer to instances that are
incorrectly forecasted as "COVID" when they are really classified as "Normal." In this particular case, there
are four instances that fall under this category. False negatives (FN) refer to instances in which the prediction
is classified as "Normal," when in reality, it should have been classified as "COVID." In this particular case,
there were three instances of FN.

Figure 2. ROC curve Figure 3. Confusion matrix

Int J Artif Intell, Vol. 14, No. 4, August 2025: 3192-3200

Int J Artif Intell ISSN: 2252-8938  3197

The precision-recall curve represented in Figure 4 that visualizes the trade-off between accuracy and
recall for various categorization thresholds. It is also known as the precision-recall curve. Predictions are
produced using a confidence score or a probability value in many classification algorithms. A threshold is
then used to decide whether a prediction is deemed positive or negative. The curve is produced by making
small, incremental changes to a predetermined threshold that is used for identifying cases. When the
threshold is adjusted, the number of TP, FP, TN, and FN also changes. This, in turn, has an impact on the
accuracy and recall of the test. The classifier has a tendency to produce fewer positive predictions when the
threshold is set extremely high, which leads to high accuracy but perhaps reduced recall. When the threshold
is low, on the other hand, more cases are projected as positive, which increases recall but may result in
decreased accuracy.
Figure 5 is F1 score evolution plot which is a statistic that combines accuracy and recall into a single
number. This value may be used to evaluate performance. It is especially helpful in situations in which you
wish to strike a compromise between the competing values of accuracy and recall. The formula that is used to
determine the F1 score is as follows:

𝐹1 𝑆𝑐𝑜𝑟𝑒 = 2 × (𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 × 𝑅𝑒𝑐𝑎𝑙𝑙) / (𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 + 𝑅𝑒𝑐𝑎𝑙𝑙)

The multiple iterations or different implementations of the categorization model are shown along the x-axis
of the Figure 5. Each iteration is a separate effort to enhance the model's functionality in a variety of ways,
and these improvements are represented as discrete attempts. The F1 score that the model managed to attain
during each iteration is shown along the y-axis. The F1 score is intended to be a measure of the degree to
which the model is able to strike a balance between recall and accuracy. When compared to previous scores,
a higher F1 score implies improved overall performance. The performance metrics of the proposed model is
reported in Table 1.

Figure 4. Precision vs recall curve Figure 5. F1 score evolution plot

Table 1. Performance metrics of the proposed model

Label Precession Recall F1 score
Normal 0.98 0.97 0.98
COVID 0.88 0.94 0.91
Accuracy 0.96

The performance measures that have been presented in Table 1, the efficiency of the model with
regard to these important metrics is shown for two separate classes, namely "Normal" and "COVID." An
accuracy value of 0.96 is shown by the model when applied to the "Normal" class. This shows that a relatively
high percentage of correct positive predictions were made in comparison to the total number of positive
predictions made. The fact that the model has a recall value of 0.97 indicates that it is capable of identifying a
significant proportion of occurrences that are really positive. A score of 0.98 for the equivalent F1 indicates a
balanced performance in terms of both accuracy and recall. In a similar way a precision value of 0.88 for the
"COVID" class indicates a satisfactory degree of accuracy in the identification of affirmative cases. The fact
that the model was successful in capturing a significant percentage of examples when the hypothesis was
correct is shown by the recall value of 0.94. The fact that this specific class has an F1 score of 0.91 indicates
that there is a healthy balance existing between accuracy and recall. Comparative results as shown in Table 2.

A deep learning-based framework for automatic detection COVID-19 using chest … (Sivanagireddy Kalli)
3198  ISSN: 2252-8938

Table 2. Comparative results

Model Accuracy (%)
AlexNet 87
VGG16 89
ResNet50 94
Proposed DenseNet model 96

5. CONCLUSION
In conclusion, a classification model was developed in our study to identify COVID-19 cases based
on medical image analysis. A dataset comprising both "Normal" and "COVID" cases was utilized, and
machine learning techniques were employed to train and evaluate the model's performance. Promising results
in terms of accuracy, precision, recall, and F1 score were exhibited by the model. Specifically, an accuracy of
96% was achieved for the "Normal" class, indicating a high percentage of correctly predicted negative cases.
The recall of 97% suggested that actual negative cases were proficiently identified, and a well-balanced
performance between accuracy and recall was indicated by the F1 score of 98%. For the "COVID" class, a
precision of 88% was attained by the model, demonstrating a satisfactory level of accuracy in identifying
positive cases. The recall of 94% signified the model's capability to capture a significant proportion of actual
positive cases, and the F1 score of 91% confirmed a healthy balance between accuracy and recall for this class.
Comparison to other existing models, including AlexNet, VGG16, and ResNet50, was conducted to evaluate
our model's performance. Our proposed DenseNet model outperformed these models with an accuracy of 96%,
showcasing its effectiveness in the detection of COVID-19 from medical images.

ACKNOWLEDGMENTS
The authors wish to thanks ahead of time the Department of Electronics and Communications
Engineering and Department of Computer Science and Engineering, Sridevi Women Engineering College,
Hyderabad, to assist them in their technical advice and research amenities. We also acknowledge, Dr. A.
Narmada, who provided useful comments during model validation.

FUNDING INFORMATION
The study is partially sponsored by the Internal Quality Assurance Cell (IQAC) scheme of the
Sridevi Women Engineering College in terms of infrastructure and research facilities. The produced work
was not funded by any agency.

AUTHOR CONTRIBUTIONS STATEMENT

This journal uses the Contributor Roles Taxonomy (CRediT) to recognize individual author
contributions, reduce authorship disputes, and facilitate collaboration.

Name of Author C M So Va Fo I R D O E Vi Su P Fu
Sivanagireddy Kalli ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Bukka Narendra Kumar ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Saggurthi Jagadeesh ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Kushagari Chandramouli ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Ravi Kumar

C : Conceptualization I : Investigation Vi : Visualization

M : Methodology R : Resources Su : Supervision
So : Software D : Data Curation P : Project administration
Va : Validation O : Writing - Original Draft Fu : Funding acquisition
Fo : Formal analysis E : Writing - Review & Editing

CONFLICT OF INTEREST STATEMENT

The authors indicate that they do not have any known competing financial interests or personal
relationships that might have seemed to affect the work reported in this paper. There is no conflict of interest
declared by authors.

Int J Artif Intell, Vol. 14, No. 4, August 2025: 3192-3200

Int J Artif Intell ISSN: 2252-8938  3199

INFORMED CONSENT
All the people included in this work have signed an informed consent with us. Participants were
made to understand what the research involved and their consent was given as per the ethical requirements.

ETHICAL APPROVAL
All the research working with human subjects followed all the necessary national regulations and
institutional policies, as well as followed the principles of the Helsinki Declaration. The Institutional Review
Board (IRB) of Sridevi Women Engineering College, Hyderabad, Telangana, India. The study is not linked
to human subjects or animal research and thus does not need ethical scrutiny.

DATA AVAILABILITY
The data underlying the findings of the present study can be requested by addressing the
corresponding author with a reasonable request. The data that was utilized in the research is not openly
available or confidential. The data that was utilized in the research is not openly available as per institutional
confidentiality policy but can be shared as per reasonable request addressed to the corresponding author.

REFERENCES
[1] WHO, “Clinical management of severe acute respiratory infection when novel coronavirus (2019-nCoV) infection is suspected:
interim guidance,” World Health Organization, 2020.
[2] H. Shi et al., “Radiological findings from 81 patients with COVID-19 pneumonia in Wuhan, China: a descriptive study,” The
Lancet Infectious Diseases, vol. 20, no. 4, pp. 425–434, 2020, doi: 10.1016/S1473-3099(20)30086-4.
[3] A. W. Salehi, P. Baglat, and G. Gupta, “Review on machine and deep learning models for the detection and prediction of
Coronavirus,” Materials Today: Proceedings, vol. 33, pp. 3896–3901, 2020, doi: 10.1016/j.matpr.2020.06.245.
[4] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp. 436–444, 2015, doi: 10.1038/nature14539.
[5] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE Computer
Society Conference on Computer Vision and Pattern Recognition, 2016, pp. 770–778, doi: 10.1109/CVPR.2016.90.
[6] I. D. Apostolopoulos and T. A. Mpesiana, “COVID-19: automatic detection from X-ray images utilizing transfer learning with
convolutional neural networks,” Physical and Engineering Sciences in Medicine, vol. 43, no. 2, pp. 635–640, 2020, doi:
10.1007/s13246-020-00865-4.
[7] T. Ozturk, M. Talo, E. A. Yildirim, U. B. Baloglu, O. Yildirim, and U. Rajendra Acharya, “Automated detection of COVID-19
cases using deep neural networks with X-ray images,” Computers in Biology and Medicine, vol. 121, 2020, doi:
10.1016/j.compbiomed.2020.103792.
[8] L. Wang, Z. Q. Lin, and A. Wong, “COVID-Net: a tailored deep convolutional neural network design for detection of COVID-19
cases from chest X-ray images,” Scientific Reports, vol. 10, no. 1, Nov. 2020, doi: 10.1038/s41598-020-76550-z.
[9] A. Jaiswal, N. Gianchandani, D. Singh, V. Kumar, and M. Kaur, “Classification of the COVID-19 infected patients using
DenseNet201 based deep transfer learning,” Journal of Biomolecular Structure and Dynamics, vol. 39, no. 15, pp. 5682–5689,
2021, doi: 10.1080/07391102.2020.1788642.
[10] Y. Song et al., “Deep learning enables accurate diagnosis of novel coronavirus (COVID-19) with CT images,” IEEE/ACM Transactions
on Computational Biology and Bioinformatics, vol. 18, no. 6, pp. 2775–2780, 2021, doi: 10.1109/TCBB.2021.3065361.
[11] L. Yan and X. Liming, “Coronavirus disease 2019 (COVID-19): role of chest CT in diagnosis and management,” American
Journal of Roentgenology, vol. 214, no. 6, pp. 1280–1286, 2020.
[12] P. Zhou et al., “A pneumonia outbreak associated with a new coronavirus of probable bat origin,” Nature, vol. 579, no. 7798,
pp. 270–273, 2020, doi: 10.1038/s41586-020-2012-7.
[13] L. Min, C. Qiang, and Y. Shuicheng, “Network in network,” in 2nd International Conference on Learning Representations, ICLR
2014 - Conference Track Proceedings, 2014.
[14] H. Nishiura, N. M. Linton, and A. R. Akhmetzhanov, “Serial interval of novel coronavirus (COVID-19) infections,” International
Journal of Infectious Diseases, vol. 93, pp. 284–286, 2020, doi: 10.1016/j.ijid.2020.02.060.
[15] M. Turkoglu, “COVIDetectioNet: COVID-19 diagnosis system based on X-ray images using features selected from pre-learned
deep features ensemble,” Applied Intelligence, vol. 51, no. 3, pp. 1213–1226, 2021.
[16] A. A. Ardakani, A. R. Kanafi, U. R. Acharya, N. Khadem, and A. Mohammadi, “Application of deep learning technique to
manage COVID-19 in routine clinical practice using CT images: results of 10 convolutional neural networks,” Computers in
Biology and Medicine, vol. 121, 2020, doi: 10.1016/j.compbiomed.2020.103795.
[17] L. Brunese, F. Mercaldo, A. Reginelli, and A. Santone, “Explainable deep learning for pulmonary disease and coronavirus COVID-19
detection from X-rays,” Computer methods and programs in biomedicine, vol. 196, 2020, doi: 10.1016/j.cmpb.2020.105608.
[18] Y. Chen, Q. Liu, and D. Guo, “Emerging coronaviruses: Genome structure, replication, and pathogenesis,” Journal of Medical
Virology, vol. 92, no. 4, pp. 418–423, 2020, doi: 10.1002/jmv.25681.
[19] S. B. Stoecklin et al., “First cases of coronavirus disease 2019 (COVID-19) in France: surveillance, investigations and control
measures, January 2020,” Eurosurveillance, vol. 25, no. 6, 2020, doi: 10.2807/1560-7917.ES.2020.25.6.2000094.
[20] D. P. Fan et al., “Inf-Net: automatic COVID-19 lung infection segmentation from CT images,” IEEE Transactions on Medical
Imaging, vol. 39, no. 8, pp. 2626–2637, 2020, doi: 10.1109/TMI.2020.2996645.
[21] R. M. Pereira, D. Bertolini, L. O. Teixeira, C. N. Silla, and Y. M. G. Costa, “COVID-19 identification in chest X-ray images on
flat and hierarchical classification scenarios,” Computer Methods and Programs in Biomedicine, vol. 194, 2020, doi:
10.1016/j.cmpb.2020.105532.
[22] EusoMII, “A European initiative for automated diagnosis and quantitative analysis of COVID-19 on imaging,” EusoMII, 2020. [Online].
Available: https://www.eusomii.org/a-european-initiative-for-automated-diagnosis-and-quantitative-analysis-of-covid-19-on-imaging/
[23] N. Zhang et al., “Recent advances in the detection of respiratory virus infection in humans,” Journal of Medical Virology,
vol. 92, no. 4, pp. 408–417, 2020, doi: 10.1002/jmv.25674.

A deep learning-based framework for automatic detection COVID-19 using chest … (Sivanagireddy Kalli)
3200  ISSN: 2252-8938

[24] RSNA, “RSNA announces COVID-19 imaging data repository,” Radiological Society of North America, 2021. [Online].
Available: https://www.rsna.org/news/2020/march/covid-19-imaging-data-repository
[25] V. M. Corman et al., “Detection of 2019 novel coronavirus (2019-nCoV) by real-time RT-PCR,” Eurosurveillance, vol. 25, no. 3,
2020, doi: 10.2807/1560-7917.ES.2020.25.3.2000045.
[26] J. Wang and L. Perez, “The effectiveness of data augmentation in image classification using deep learning,” arXiv-Computer
Science, pp. 1-8, 2017.
[27] P. Yadav, N. Menon, V. Ravi, and S. Vishvanathan, “Lung-GANs: unsupervised representation learning for lung disease
classification using chest CT and X-ray images,” IEEE Transactions on Engineering Management, vol. 70, no. 8, pp. 2774–2786,
2023, doi: 10.1109/TEM.2021.3103334.
[28] J. Deng, W. Dong, R. Socher, L. J. Li, K. Li, and L. Fei-Fei, “ImageNet: a large-scale hierarchical image database,” in 2009 IEEE
Conference on Computer Vision and Pattern Recognition, CVPR 2009, 2009, pp. 248–255, doi: 10.1109/CVPR.2009.5206848.
[29] A. Arunachalam, V. Ravi, V. Acharya, and T. D. Pham, “Toward data-model-agnostic autonomous machine-generated data
labeling and annotation platform: COVID-19 autoannotation use case,” IEEE Transactions on Engineering Management, vol. 70,
no. 8, pp. 2695–2706, 2023, doi: 10.1109/TEM.2021.3094544.
[30] A. Ben-Cohen, E. Klang, M. M. Amitai, J. Goldberger, and H. Greenspan, “Anatomical data augmentation for CNN based pixel-wise
classification,” in 2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018), 2018, pp. 1096–1099, doi:
10.1109/ISBI.2018.8363762.
[31] D. Mukhtorov, M. Rakhmonova, S. Muksimova, and Y. I. Cho, “Endoscopic image classification based on explainable deep
learning,” Sensors, vol. 23, no. 6, 2023, doi: 10.3390/s23063176.
[32] C. I. Paules, H. D. Marston, and A. S. Fauci, “Coronavirus infections more than just the common cold,” JAMA-Journal of the
American Medical Association, vol. 323, no. 8, pp. 707–708, 2020, doi: 10.1001/jama.2020.0757.

BIOGRAPHIES OF AUTHORS

Sivanagireddy Kalli is presently a Professor of Electronics and Communication

Engineering. He did his Ph.D. degree in Electronics and Communication Engineering from
JNTU Hyderabad in 2019. He is having more than 22 years of teaching and 8 years of research
experience. He has a total of 60 research publications in international journals and 9 patents. His
current research areas are artificial intelligence, machine learning, and deep learning. He is a
member of IEEE. He can be contacted at email: sivanagireddykalli@gmail.com.

Bukka Narendra Kumar is presently a Professor of Computer Science and

Engineering and HoD. He did his Ph.D. degree in Computer Science and Engineering from
JNTU Hyderabad in 2019. He is having more than 25 years of teaching and 8 years of research
experience. He has a total of 20 research publications in international journals and 4 patents. His
current research areas are information security, artificial intelligence, machine learning, and deep
learning. He is a life member of MISTE. He can be contacted at email: bnkphd@gmail.com.

Saggurthi Jagadeesh is presently Professor of Electronics and Communication

Engineering. He did his Ph.D. degree in Electronics and Communication Engineering from
JNTU Hyderabad in 2019. He is having more than 25 years of Teaching and 8 years of research
experience. He has a total of 40 research publications in international journals and 9 patents. His
current research areas are artificial intelligence, machine learning, and deep learning. He is a
member of IEEE. He can be contacted at email: jaaga.ssjec@gmail.com.

Kushagari Chandramouli Ravi Kumar is presently Professor of Computer

Science and Engineering. He did his Ph.D. degree in Computer Science and Engineering from
JNTU Hyderabad in 2019. He is having more than 29 years of teaching and 8 years of research
experience. He has a total of 20 research publications in international journals and 3 patents. His
current research areas are artificial intelligence, machine learning, and deep learning. He is a
member of MISTE. He can be contacted at email: kcravikunar1971@gmail.com.

Int J Artif Intell, Vol. 14, No. 4, August 2025: 3192-3200