Genetic Algorithm Based Optimization of Convolutional Neural

Network for Respiratory Disease Detection

Vishwachetan D

, Nandini S B

Pranjal Shrivastava and Nihal Jahagirdar

Department of Computer Science and Engineering, Ramaiah Institute of Technology, Bangalore, India

Keywords: X-Ray Imaging, AI, Xception Neural Network, Pulmonary Disease, Genetic Algorithm.

Abstract:

The pandemic Covid 19 in the year 2019 highlighted the need for advanced diagnostic

methodologies to address a spectrum of pulmonary diseases. Although the major method of

COVID-19 detection is still conventional PCR testing, the combination of AI and X-ray imaging

presents a promising path toward a thorough diagnosis of pulmonary illness. Here, we provide a

new optimization framework based on the Xception neural network architecture and Genetic

Algorithm (GA) for precise pulmonary disease detection from X-ray pictures, including

coronavirus and pneumonis (viral, bacterial). By utilising deep learning and convolutional neural

networks, the main aim of this paper to improve the accuracy and efficiency of diagnosis. Using

GA, we explore the vast design space of deep CNN architectures, encompassing parameters such

as network depth, layer count, and type. Utilising an extensive dataset of X-ray pictures, the

suggested Xception-based neural network is rigorously assessed repeatedly through GA-driven

optimization. The result highlight how well the improved model distinguishes lung disorders

achieved with AI-driven approaches.

1 INTRODUCTION

The technology based on deep learning algorithms

has transformed traditional medical image diagnosis

and prognosis in recent years. X-ray imaging is of the

most readily available and used method for

diagnosing lung conditions like pneumonia, whether

bacterial or viral, as well as emerging threats like

COVID-19. The use of deep neural networks (DNNs)

for automated classification and detection tasks has

seen significant advancements, delivering remarkable

accuracy across various benchmarks. Among the

notable CNN architectures is Xception, introduced in

2017 by François Chollet, known for its exceptional

image recognition capabilities. Its innovative depth-

wise separable convolutions enhance effective

feature learning and extraction, making it a strong

candidate for medical image analysis where accuracy

and computational efficiency are paramount.

However, fine-tuning Xception's architecture and

parameters to meet specific medical imaging

https://orcid.org/0009-0009-9676-8279

https://orcid.org/0009-0001-5708-2954

requirements remains a challenge. This study

suggests a way to enhance the detection of pulmonary

diseases, including COVID-19, viral, and bacterial

pneumonia, in X-ray images. The approach involves

a hybrid methodology that combines the optimization

capabilities of genetic algorithms (GAs) with

Xception's properties. Genetic algorithms, inspired

by natural selection, serve as a powerful tool to

identify and build optimal DNN configurations. This

research seeks to determine whether using genetic

algorithms to refine the Xception model's architecture

and hyperparameters—specifically for lung disease

detection from X-ray images—is effective. The

objective is to improve the precision and robustness

of disease diagnosis by iteratively adjusting neural

network architectures using genetic algorithms,

facilitating early identification and timely diagnosis

or treatment. The optimized Xception-based DNN

model will be evaluated on benchmark datasets

through extensive experiments, including cases of

COVID-19, viral pneumonia, bacterial pneumonia,

804

D, V., S B, N., Shrivastava, P. and Jahagirdar, N.

Genetic Algorithm Based Optimization of Convolutional Neural Network for Respiratory Disease Detection.

DOI: 10.5220/0013733200004664

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 3, pages 804-810

ISBN: 978-989-758-763-4

and healthy controls. The aim of this study is to

advance the creation of reliable and efficient

automated pulmonary disease diagnosis tools,

supporting clinical decisions by healthcare

professionals and ultimately improving patient

outcomes.

2 LITERATURE REVIEW

Over the years, several techniques have been

suggested for medical image analysis, shifting from

traditional feature-based approaches to advanced

machine learning techniques. Early investigations in

medical image analysis focused on basic image

processing techniques like thresholding,

morphological operations, and edge detection. While

these methods laid the groundwork for future

research, they often fell short in meeting the accuracy

and consistency required in clinical practice.

Apostolopoulos and Mpesiana (Apostolopoulos,

Mpesiana, et al. 2020) fine-tuned Convolutional

Neural Networks (CNNs) for the automatic detection

of COVID-19 from X-ray images, demonstrating that

pre-trained models can enhance diagnostic accuracy.

Similarly, Duran-Lopez et al. (Duran-Lopez,

ominguez-Morales, et al. 2020) proposed COVID-

XNet, a deep learning model designed to diagnose

and localize COVID-19 in chest X-rays, aiming to

improve both detection accuracy and efficiency.

Sethy and Behera (Sethy, Behera, et al. 2020)

investigated deep learning potential in medical

imaging, using neural networks to extract features for

the identification of COVID-19 in X- rays images.

This study highlighted how deep learning techniques

can streamline the analysis of X-ray data, offering a

reliable solution for disease diagnosis and reducing

unnecessary examinations.Narin et al. (Narin, Kaya,

et al. 2021) applied various deep learning models for

classification of coronavirus and normal cases, with

their ResNet50 model achieving 98.0% accuracy in

the best-case scenario. Zhang et al. (Zhang, Xie, et al.

2020) introduced another ResNet-based model that

achieved an AUC of 0.952, effectively highlighting

areas affected by pneumoniavari using Grad-CAM.

Wang et al. (Wang, Lin, et al. 2020) proposed a deep

CNN for classifying viral and bacterial infections and

normal cases, achieving 83.5% accuracy.Image

segmentation has also played a critical role in

COVID-19 applications, including diagnostics

(Chen, et al. 2019), (Wang, et al. 2021), (Jin, et al.

2020), (Song, et al. 2021). For example, Li et al. (Li,

et al. 2020) used a U-Net architecture to segmentin

lung images to differentiate COVID-19 from

pneumonia acquired from the community using CT

scans of the chest region. Jin et al. (Jin, et al. 2020)

developed an AI system for rapid COVID-19

detection, where segmented CT slices serve as input

for the classification model.Segmentation techniques

also prove valuable in quantification tasks within

medical applications (Jin, et al. 2020), (Shan, et al.

2021). A new model, XcepCovidNet, was introduced

to identify features in X-rays of the chest region,

utilizing transfer learning combined with

hyperparameter tuning to address limitations in the

training dataset (Juneja, Kumar, et al. 2024). Beyond

X-rays, recent studies have turned to CT scans for the

same purpose. For instance, Khan et al. (Khan, Shah,

et al. 2020) developed CoroNet, a classification

system consisting of four classes for COVID-19,

achieving accuracies of 89.6% and 95% for chest X-

ray (CXR) and CT scanned images, respectively.

COVNet, designed by Li et al. (Li, et al. 2020), was

based on ResNet50 and trained on a dataset of 4,356

images of CT scans of the chest region. Lastly,

Joloudari et al. (Joloudariet, et al. 2023) proposed a

deep learning-based global feature extractor for

COVID-19 detection, further contributing to the

research on using deep learning in medical image

analysis.

3 PROPOSED METHOD

The suggested model in this paper makes use of three

fundamental algorithms. The deep convolutional

neural network Xception model is used for the

detection of respiratory disease in lung X-ray images.

The Genetic algorithm is then used to tune the

hyperparameters of the Xception model to achieve the

best possible architecture. This section explains the

Genetic Algorithm, CNN, Xception model and finally

the model suggested.

3.1 CNN

CNNs are a class of deep learning models for

processing structured grid data, notably images and

videos. Convolutional, pooling, and fully connected

layers are how CNNs work to extract hierarchical

characteristics from input data.Components of a

CNN: Convolutional Layers: These layers are made

up of filters, sometimes known as kernels, that

execute convolutions by sliding across the input data.

Each filter specializes in detecting specific features,

such as edges or textures, by capturing spatial

correlations. Activation functions such as ReLU

introduce non-linearity. Pooling Layers:

Genetic Algorithm Based Optimization of Convolutional Neural Network for Respiratory Disease Detection

805

Convolutional layer feature maps are down sampled

by pooling layers, which reduces spatial dimensions

without losing important information. For example,

max pooling selects the greatest value within local

regions, effectively shrinking feature map sizes.

Feature maps are converted into vectors and run

through one or more fully connected layers following

a number of convolutional and pooling layers.The

layers here handle classification or regression tasks

by learning intricate relationships between extracted

features and target labels. CNN architectures may

incorporate additional elements such as dropout

layers for regularization, batch normalization layers

for accelerated convergence, and skip connections for

improved gradient flow during training. Training

CNNs involves optimizing parameters (weights and

biases) using gradient-based optimization algorithms

like SGD or its variants. During training, the network

minimizes a loss function, quantifying the disparity

between predicted outputs and ground truth labels.

3.2 Xception

Xception is a CNN architecture innovated by

François Chollet, renowned for his contribution to the

Keras deep learning library. Termed as "Extreme

Inception," Xception builds upon the foundational

concepts of the Inception architecture while

introducing notable advancements. Central to

Xception's design is the utilization of depth wise

separable convolutions, a variant of conventional

convolutional operations. This methodology

effectively segregates spatial and channel-wise

convolutions into distinct processes, resulting in a

significant reduction in both parameters and

computational complexity compared to conventional

convolutions. Consequently, Xception achieves

enhanced efficiency and model lightweightness. The

architecture of Xception heavily draws from the

Inception modules featured in the Inception v3

model. However, Xception distinguishes itself by

replacing conventional convolutions within these

modules with depth wise separable convolutions.

This architectural refinement facilitates an improved

utilization of computational resources, ensuring the

modeling of intricate patterns and relationships across

various scales. One notable advantage of Xception

lies in its capability to capture both local and global

dependencies within input data. The decomposition

of the convolution operation into spatial and channel-

wise components allows Xception to effectively

model complex structures and correlations present in

the data. Furthermore, Xception's architecture boasts

expedited training and inference times, surpassing

preceding CNN architectures. This attribute renders

Xception particularly suitable for applications

characterized by resource-constrained environments,

where computational efficiency is paramount.

3.3 Genetic Algorithm

Genetic Algorithms (GAs) are commonly applied

to optimization problems, such as tuning

hyperparameters in machine learning. A population

of potential solutions, known as people or

chromosomes, is used by a genetic algorithm to solve

problems. Each chromosome is a potential solution.

In the context of hyperparameter tuning, these

solutions usually correspond to different sets of

hyperparameters for a machine learning model. As

the algorithm runs through a sequence of stages called

generations, selection, crossover, and mutation are

applied to create a fresh set of potential solutions.

3.3.1 Initialization

The algorithm begins by generating an initial

population of chromosomes, typically done

throughrandomness or certain heuristics.

3.3.2 Evaluation

Each chromosome is assessed according to its fitness,

which measures how effectively the solution

performs the given optimization task. In

hyperparameter tuning, the fitness is determined by

how well the machine learning model performs when

trained with the hyperparameters encoded in the

chromosome.

3.3.3 Selection

Greater fitness values in chromosomes are preferable

for reproduction, similar to the concept of "survival

of the fittest."

3.3.4 Crossover

Selected chromosomes are paired to produce

offspring through crossover or recombination. This

process involves mixing genetic information from

parent chromosomes to create new solutions. In

hyperparameter tuning, crossover allows for the

exploration of different hyperparameter

combinations.

In hyperparameter tuning, where finding the ideal

set of hyperparametersfor a machine learning model is

the goal, GAs provide an effective way to navigate the

large search space. Initially, a population of potential

INCOFT 2025 - International Conference on Futuristic Technology

806

solutions (chromosomes) is created either randomly or

using heuristics. Each chromosome encodes some

hyperparameters for the machine learning model,

which might include factors like learning rates,

regularization strengths, or network

architectures.Next, the fitness of each chromosome is

assessed by training and testing the machine learning

model using the hyperparameters it contains.

Performance metrics, such as accuracy or loss, are

used to assess fitness. Selection methods like

tournament or roulette wheel selection are then

applied to choose chromosomes to be reproduced

based on their fitness, favoring those with greater

values.Crossover and mutation are triedon these

selected chromosomes to generate new offspring.

Crossover mixes genetic information from parent

chromosomes, enabling the exploration of new

hyperparameter combinations. Mutation introduces

random changes, maintaining diversity in the

population and preventing early convergence to

suboptimal solutions. The offspring replace the

previous generation, with fitter individuals more

likely to survive.Until a termination requirement is

satisfied, for example, by reaching a certain number of

generations or attaining adequate performance, this

iterative process keeps going. Through this cycle of

selection, crossover, and mutation, genetic algorithms

efficiently search the hyperparameter space, slowly

converging toward optimal or near-optimal

configurations that enhance the ML model's

performance.

Figure 1: Flowchart of proposed model

This research employs Genetic Algorithm (GA) in

conjunction with the Xception architecture, a

sophisticated Convolutional Neural Network (CNN).

First, a deep Xception network is constructed

utilizing parameters computed from the Genetic

Algorithm. Subsequently, these deep neural

networks undergo training and evaluation using a

dataset aimed at discerning COVID-19 presence in

individuals. Each network's performance is assessed

based on its error rate, with lower error rates

indicative of more desirable solutions. These

evaluated solutions undergo further refinement via

iterations of the GA algorithm. With each iteration,

novel networks are generated, leading to

progressively improved outcomes. The rationale

behind selecting the Xception model stems from its

proven efficacy in prior research endeavours,

consistently yielding commendable results. Its

selection is particularly apt given its tailored focus on

COVID-19 detection, aligning closely with the

objectives of this study. Figure 1 explains the flowof

the proposed model.

4 IMPLEMENTATION

4.1 Dataset for Respiratory Disease

Detection Training

The dataset for respiratory disease detection training

encompasses four primary classes: COVID, normal

lung conditions, bacterial pneumonia, and viral

pneumonia. Initially split into training, testing, and

validation sets, the dataset underwent meticulous

cleaning due to the presence of noise, including

random letters and unnecessary watermarks, ensuring

data integrity and reliability. With close to 1400

images solely for training purposes, augmentation

techniques were employed to expand the dataset size,

enhancing the model's ability to generalize and learn

diverse patterns. This comprehensive dataset,

meticulously curated and augmented, serves as a

robust foundation for training and evaluating deep

learning models aimed at accurate and effective

respiratory disease detection and classification.

4.2 Model Implementation

For implementing the proposed model, after

initializing the population, the evaluate_population

function is called to train and evaluate each candidate

solution (CNN) on the training and validation

datasets. This step involves training the CNN model

with the training data, then evaluating its execution

Genetic Algorithm Based Optimization of Convolutional Neural Network for Respiratory Disease Detection

807

using the validation data. The fitness scores are

computed based on the performance metrics obtained

during evaluation. Within the loop that iterates

through generations, the training of CNN models is

implicitly done during the evaluation step, as part of

the evaluate_population function. The CNN models

are trained with the training data before their

performance is evaluated on the validation data. The

training process typically involves using the training

data on the CNN model, computing the loss, and then

modifying the model's parameters. Algorithms for

optimization such as SGD is used. The performance

of each CNN model on the validation dataset is then

evaluated using metrics like accuracy, loss, or other

relevant measures, and these evaluations are used to

compute the fitness scores. The process continues

until a termination condition is reached, such as

reaching a set number of generations.

5 RESULTS

Accuracy and categorical cross entropy loss were

theperformance metrics used to find thebest

architecture discovered by Genetic algorithm. One

frequent metric used to assess a classification model's

performance is accuracy. Out of all the anticipated

classifications, it calculates the percentage of accurate

classifications. Accuracy is mathematically

calculated as shown in Figure 2.

The categorical cross-entropy loss function given

in Equation 1serves as a pivotal evaluation metric in

various machine learning tasks, particularly in

classification problems where the output is

represented in a categorical format. This metric

quantifies the disparity between the true distribution

of class labels and the predicted probabilities assigned

by the model. By computing the logarithmic

difference between the predicted probabilities and the

actual class labels across all categories, the

categorical cross-entropy loss penalises deviations

from the true distribution, effectively guiding the

model towards better classification performance. Its

formulation makes it particularly suited for multi-

class classification tasks, providing a continuous,

differentiable measure of the model's performance

that can be optimised through gradient descent

methods.

Loss = 

∑

𝑦









⋅𝑙𝑜𝑔𝑦



(1)

where, loss is the categorical cross-entropy loss, n

is the output size, y is the correct probability

distribution of class labels (one-hot encoded) and y^

is the estimated probability distribution of class

labels.

The accuracy metric demonstrates a notable

enhancement, with a consistent increase of 2-3%,

indicating improved model performance in correctly

classifying data points.

Figure 2: Accuracy obtained over 80 Epochs

The cross-entropy loss in Figure 3 exhibits a

significant improvement, with a remarkable decrease

of 30%, reflecting the model's enhanced ability to

minimize discrepancies between predicted and true

class probabilities. These advancements, coupled

with the absence of significant spikes in loss, suggest

the efficacy of hyperparameter tuning and the

refinement of the dataset, contributing to a more

stable and robust model performance with reduced

noise interference.

Figure 3: Categorical Cross Entropy Loss over 80 Epochs

The model trained was then tested across various

images of lungs from the dataset to evaluate its

performance. The accuracy on the test dataset was

INCOFT 2025 - International Conference on Futuristic Technology

808

found out to be 0.79 and after optimization using GA,

an accuracy of 0.85 was obtained. Categorical cross

entropy loss of 0.79 was also reduced to 0.44 upon

optimization by GA.

Figure 4: Predictions obtained for various classes

Figure 4 shows the results of classification along

with their predicted conditions obtained from the

trained model. The confusion matrices shown in

Figure 5 and Figure 6 depict the accuracy of

predictions of the model before and after

optimizations using the Genetic Algorithm.

Figure 5: Confusion Matrix on test dataset before GA

optimization

Figure 6 Confusion Matrix on test dataset after GA

optimization

6 CONCLUSIONS

In conclusion, the enhancements made to the model

have yielded significantly improved stability and

robustness, as evidenced by the obtained results.

Notably, there is a consistent enhancement in

accuracy ranging between 2-3%, indicating the

effectiveness of the implemented changes. Equally

significant is the remarkable 30% decrease in

Validation Loss, underscoring the model's improved

generalisation capability. It's important to note that

while training loss measures the performance of the

model during the training phase, validation loss

provides insight into how well the model generalises

to unseen data, making it a crucial metric in assessing

real-world performance. The fact that both training

and validation accuracy and losses closely match

underscores the absence of overfitting, signifying that

the model has learned to generalise well to unseen

data. This alignment between training and validation

metrics further validates the reliability and efficacy of

the model's performance. Overall, these results affirm

the success of the enhancements implemented, paving

the way for more reliable and accurate predictions in

practical applications.

7 FUTURE WORK

For future work, a larger dataset should be assembled

to encompass a more diverse set of lung diseases,

potentially enhancing the model's ability to detect and

classify a broader range of conditions. Additionally,

fine-tuning the existing architecture with progressive

Genetic Algorithm Based Optimization of Convolutional Neural Network for Respiratory Disease Detection

809

techniques such as transfer learning or ensemble

methods could potentially elevate the model's

performance to even greater heights.

REFERENCES

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, Apr. 2020, doi:

10.1007/s13246-020-00865-4. Available:

https://doi.org/10.1007/s13246-020-00865-4

L. Duran-Lopez, J. P. Dominguez-Morales, J. Corral-

Jaime, S. Vicente-Diaz, and A. Linares-Barranco,

“COVID-XNet: A Custom Deep Learning System to

Diagnose and Locate COVID-19 in Chest X-ray

Images,” Applied Sciences, vol. 10, no. 16, p. 5683,

Aug. 2020, doi: 10.3390/app10165683. Available:

https://doi.org/10.3390/app10165683

P. K. Sethy, S. K. Behera, P. K. Ratha, and P. Biswas,

“Detection of coronavirus Disease (COVID-19) based

on Deep Features and Support Vector Machine,”

International Journal of Mathematical Engineering and

Management Sciences, vol. 5, no. 4, pp. 643–651, Aug.

2020, doi: 10.33889/ijmems.2020.5.4.052. Available:

https://doi.org/10.33889/ijmems.2020.5.4.052

A. Narin, C. Kaya, and Z. Pamuk, “Automatic detection of

coronavirus disease (COVID-19) using X-ray images

and deep convolutional neural networks,” Pattern

Analysis and Applications, vol. 24, no. 3, pp. 1207–

1220, May 2021, doi: 10.1007/s10044-021-00984-y.

Available: https://doi.org/10.1007/s10044-021-00984-

Zhang J, Xie Y, Li Y, Shen C, Xia Y. COVID-19 Screening

on Chest X-ray Images Using Deep Learning based

Anomaly Detection. arXiv; 2020.

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. Available:

https://doi.org/10.1038/s41598-020-76550-z

O. Gozes et al., “Rapid AI Development Cycle for the

Coronavirus (COVID-19) Pandemic: Initial Results for

Automated Detection & Patient Monitoring using Deep

Learning CT Image Analysis,” arXiv (Cornell

University), Jan. 2020, doi:

10.48550/arxiv.2003.05037. Available:

https://arxiv.org/abs/2003.05037

L. Li et al., “Using Artificial Intelligence to Detect COVID-

19 and Community-acquired Pneumonia Based on

Pulmonary CT: Evaluation of the Diagnostic

Accuracy,” Radiology, vol. 296, no. 2, pp. E65–E71,

Aug. 2020, doi: 10.1148/radiol.2020200905.

Available: https://doi.org/10.1148/radiol.2020200905

J. Chen et al., “Deep learning-based model for detecting

2019 novel coronavirus pneumonia on high-resolution

computed tomography,” Scientific Reports, vol. 10, no.

1, Nov. 2020, doi: 10.1038/s41598-020-76282-0.

Available: https://www.nature.com/articles/s41598-

020-76282-0

B. Wang et al., “AI-assisted CT imaging analysis for

COVID-19 screening: Building and deploying a

medical AI system,” Applied Soft Computing, vol. 98,

p. 106897, Jan. 2021, doi: 10.1016/j.asoc.2020.106897.

Available: https://doi.org/10.1016/j.asoc.2020.106897

C. Jin et al., “Development and Evaluation of an AI System

for COVID-19 Diagnosis,” medRxiv (Cold Spring

Harbor Laboratory), Mar. 2020, doi:

10.1101/2020.03.20.20039834. Available:

https://doi.org/10.1101/2020.03.20.20039834

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, Nov.

2021, doi: 10.1109/tcbb.2021.3065361. Available:

https://doi.org/10.1109/tcbb.2021.3065361

Y. Cao et al., “Longitudinal Assessment of COVID-19

Using a Deep Learning–based Quantitative CT

Pipeline: Illustration of Two Cases,” Radiology

Cardiothoracic Imaging, vol. 2, no. 2, p. e200082, Apr.

2020, doi: 10.1148/ryct.2020200082. Available:

https://doi.org/10.1148/ryct.2020200082

F. Shan et al., “Abnormal lung quantification in chest CT

images of COVID‐19 patients with deep learning and

its application to severity prediction,” Medical Physics,

vol. 48, no. 4, pp. 1633–1645, Mar. 2021, doi:

10.1002/mp.14609. Available:

https://doi.org/10.1002/mp.14609

A. Juneja, V. Kumar, M. Kaur, D. Singh, and H.-N. Lee,

“XcepCovidNet: deep neural networks-based COVID-

19 diagnosis,” Multimedia Tools and Applications, Jun.

2024, doi: 10.1007/s11042-024-19046-6. Available:

https://doi.org/10.1007/s11042-024-19046-6

A. I. Khan, J. L. Shah, and M. M. Bhat, “CoroNet: A deep

neural network for detection and diagnosis of COVID-

19 from chest x-ray images,” Computer Methods and

Programs in Biomedicine, vol. 196, p. 105581, Nov.

2020, doi: 10.1016/j.cmpb.2020.105581. Available:

https://doi.org/10.1016/j.cmpb.2020.105581

L. Li et al., “Using Artificial Intelligence to Detect COVID-

19 and Community-acquired Pneumonia Based on

Pulmonary CT: Evaluation of the Diagnostic

Accuracy,” Radiology, vol. 296, no. 2, pp. E65–E71,

Aug. 2020, doi: 10.1148/radiol.2020200905.

Available: https://doi.org/10.1148/radiol.2020200905

J. H. Joloudariet al., “Developing a Deep Neural Network

model for COVID-19 diagnosis based on CT scan

images,” Mathematical Biosciences & Engineering,

vol. 20, no. 9, pp. 16236–16258, Jan. 2023, doi:

10.3934/mbe.2023725. Available:

https://doi.org/10.3934/mbe.2023725

INCOFT 2025 - International Conference on Futuristic Technology

810