Pneumonia Detection Using Deep Learning

Sreevarshini Aireddy, Maruthi Sanjana, Palle Uday Kiran Goud, Mattipatti Sameer,

Myla Kalyana Ramakrishna and D. Archana

Nalla Malla Reddy Engineering College, Hyderabad, Telangana, India

Keywords: Pneumonia, Deep Learning, Convolutional Neural Networks (CNNs), VGG16, Chest X‑Ray, Web‑Based

Interface.

Abstract: Pneumonia is a serious lung infection that mainly affects children and the elderly people which requires rapid

and accurate diagnosis. There are simple and effective methods for pneumonia detection, including the deep

learning techniques, such as the convolutional neural network (CNN). As of now, such techniques have

proved potential for medical image classification. In this paper, we present a pneumonia detection model built

using the VGG16 pre-trained model of convolutional neural network architecture and trained on a labeled

dataset of the chest X-ray images. The proposed system implementation is evaluated for its performance and

also compared with the traditional methods, which show the significantly improved accuracy. This paper

mostly exhibits the potential of transfer learning and using data augmentation to improve model

generalization. This model was trained and tested on a chest X-ray image of a labeled dataset, obtaining the

accuracy of around 93%. This system is deployed with a local web-based interface, which allowed the users

to upload chest X-ray images for the real-time classification. The results show that deep learning can

significantly improve the pneumonia detection, providing a more efficient, accurate, and automated

alternative to existing diagnostic techniques.

1 INTRODUCTION

Pneumonia is an inflammatory infection of the lung

and a respiratory disease caused by bacterial, viral, or

fungal agents that affects millions of people worldwide

and has a major impact on the healthcare challenges

and mortality. Traditionally chest X-ray interpretations

are implemented by human radiologists which is time-

consuming, prone to human error, and highly

dependent on their experience. In case of any

misdiagnosis or delayed diagnosis, it may result in

serious complications, so making accurate detection

and early diagnosis is crucial.

In recent years, artificial intelligence (AI) and

deep learning have developed towards automated

diagnostic systems for medical image classification.

Convolutional Neural Networks (CNNs) have

exhibited strong performance in complex patterns in

images and obtaining high accuracy in classification

tasks. Transfer learning, which utilizes pre-trained

models of CNN on large the datasets, has been

particularly effective in medical image analysis.

Traditional methods for pneumonia detection, such as

manual interpretation of the chest X-ray images, tend

to be highly subjective as they require significant

medical experience. Other machine learning methods

such as support vector machines (SVMs) and logistic

regression have been explored variously in the past

but generally require extensive feature engineering

and have less effectiveness compared with the

complex imaging data. Other deep learning models,

such as ResNet and Inception, have also been utilized

for medical imaging, however, VGG16 particularly

has simple architecture model and is highly effective

for feature extraction. We use VGG16 to enhance

pneumonia detection accuracy while minimizing the

computational complexity.

This research mainly utilized the VGG16 model

of CNN architecture for pneumonia detection in chest

X-ray images. VGG16 is a pre-trained model which

is used in the process of feature extraction, was

performed after the additional custom layers for

classification. This model was trained on a labeled

dataset containing the chest X-ray images of normal

and pneumonia-affected to develop an automated

detection system. This research evaluates the model's

performance, helping to reduce the pressure on

radiologists and improving patient outcomes and

Aireddy, S., Sanjana, M., Goud, P. U. K., Sameer, M., Ramakrishna, M. K. and Archana, D.

Pneumonia Detection Using Deep Learning.

DOI: 10.5220/0013908400004919

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

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 4, pages

75-80

ISBN: 978-989-758-777-1

aims to automate the efficient pneumonia diagnosis

process.

2 LITERATURE REVIEW

In recent years, a number of research studies have

explored deep learning for the pneumonia detection.

Rohit Kundu, Ritacheta Das et al. (2021) proposed an

ensemble based deep learning methodology for

pneumonia detection in Chest radiographs. In this the

methodology is included of three CNN architectures

namely GoogLeNet, ResNet-18, and DenseNet-121

with a weighted average of ensemble methodology

using the precision, recall, F1-score, and AUC. This

study evaluated the algorithm using the two publicly

available datasets of pneumonia with an accuracy of

98.81% and 86.85% and also a sensitivity value of

98.80% and 87.02%. The proposed method also

exceeded the state-of-the art (SOA) methods and this

method performs better than the other ensembled

techniques. Tawsifur Rahman et al. (2020) analyzed

and used four pre-trained CNNs - AlexNet,

ResNet18, DenseNet201, and SqueezeNet. They are

categorized to identify subtypes of pneumonia

(normal vs. pneumonia, bacterial vs. viral, three-class

classification also) using transfer learning to classify

chest X-ray images. The dataset comprised of 5,247

X-rays and the methods reached the classification

accuracy of 98%, 95% and 93.3% respectively. The

study also supports using AI potentially to assist with

the rapid pneumonia diagnosis and screening.

Ayush Pant, Akshat Jain et al. (2020) by

considering the higher mortality rate of pneumonia,

this paper presented an automated deep learning

approach for the early diagnosis of pneumonia

detection by utilizing CNNs. The authors tried to

combine two CNN architectures with an ensemble

model to be allowed for the improvement of issues

with the existing methods. This study provides the

improved model with robustness and performance in

which it is to develop a diagnostic tool for the health

care professionals efficiently. Faiza M Qaimkhani,

Md G Hussain et al. (2022) studied the mostly applied

deep learning methods, which are specifically ANN,

CNN, and also the VGG19 architecture, which helps

to improve pneumonia identification accuracy

efficiently. The study focuses mainly on the early

identification, and special emphasis on the cases that

occur mostly in the children, although their aim is to

provide and serve a reliable, helpful automated

system as an aid to healthcare to mainly identify the

pneumonia cases rapidly for early treatment. Shagun

Sharma, Kalpna Guleria (2023) created a pneumonia

detection model that is mostly based on the deep

learning techniques by using the VGG-16

architecture. There is a various feature extraction

from the chest X-ray images, which are then

classified for pneumonia detection. This study aims

to support the healthcare professionals by early

diagnosis and precision of pneumonia, by improving

the health resource efficiency, and also overall

outcomes of the patient. These results show that deep

learning is a possible method for identification of

pneumonia, but, there is still a need for models that

balance the high accuracy with computational

efficiency for real-time applications.

3 METHODOLOGY

3.1 Methods

Convolutional Neural Networks: Convolutional

neural networks (CNNs), are fundamental

components of deep learning-based image

classification, including pneumonia detection in chest

X-rays. Analysis of medical imaging greatly benefits

from CNNs' ability to automatically extract

hierarchical features from input images. The

architecture is formed of different types of layers,

which include pooling layers, convolutional layers,

activation functions, and fully linked layers. The

convolutional layers use learnable filters to identify

key characteristics like edges, textures, and patterns

at various levels of abstraction. Activation functions

like ReLU (Rectified Linear Unit) introduce non-

linearity which enables the model to recognize the

complex patterns in medical imaging. Pooling layers,

such as max pooling, enhance computational

efficiency while maintaining essential information by

reducing the spatial dimensions of feature maps.

Once these features are extracted and then processed

through the fully connected layers, the model

classifies the image as either ‘Normal’ or

‘Pneumonia’. As CNNs can learn spatial structures

and identify the complex patterns which human

radiologists might miss, they perform better in

medical imaging than the traditional machine

learning techniques. The automatic feature

extraction, which reduces the need for manual feature

engineering, makes CNNs a resilient and effective

method for pneumonia detection in chest X-ray

analysis.

Pre-trained Model VGG16: The most commonly

used convolutional neural network (CNN)

architecture VGG16 was pre-trained model used on

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

the large ImageNet dataset, which includes datasets

of having millions of images in hundreds of

categories of images. VGG16 is very successful in

transfer learning in the field of medical image

analysis, because of its ability to learn rich feature

representations through pre-training models,

including the pneumonia detection. There are 16

layers that form the architecture, which includes

convolutional layers, pooling layers, and fully

connected layers. Convolutional layers mainly use

tiny 3x3 receptive fields to collect the hidden

information in photos. Instead of beginning from

scratch, we usually take the advantage of VGG16's

capacity to extract high-level spatial features and

textural information from the chest X-ray images by

utilizing it as a basis model. On top of the pre-trained

model convolutional layers, more custom layers like

the dense and dropout layers are added to improve the

model for binary classification which maintains the

important hierarchial features. This method in general

significantly decreases the training time and improves

the model accuracy because VGG16 is already pre-

trained on the broad features of images that can be

optimized for pneumonia detection. Additionally, by

freezing initial layers and training on the top layers of

the pneumonia specified data. We ensure the better

model generalization which reduce the risk of

overfitting, and improve the model's ability to detect

pneumonia in real chest X-ray images.

Model Enhancement Techniques: The pneumonia

detection model is optimized with various

preprocessing, classification, and optimization

strategies to improve its performance and

generalization. Data augmentation methods such as

rescaling, flipping and rotation are mostly utilized in

image preprocessing to introduce variations in the

training data, avoid overfitting, and enhance model

robustness. These modifications optimize the model's

ability to generalize to unseen chest X-ray images.

This model was developed for binary classification,

to classify chest X-ray images as either ‘Normal’ or

‘Pneumonia’, where it applies a sigmoid activation

function in the output layer, and to optimize strategies

which include learning rate scheduling, dropout

layers and the Adam optimizer are used to achieve

even better performance. When validation loss

gradually decreases then the learning rate

management supports stable convergence by

lowering the rate, at the same time the Adam

optimizer automatically optimizes learning rates for

improved training performance. To prevent

overfitting and improve generalization, dropout

layers acts as a regularization method by randomly

disabling neurons while training. The combination of

these methods results with high accuracy and

effective model of deep learning for pneumonia

detection.

Figure 1 shows the Normal chest X-ray and

Figure 2 shows the Pneumonia chest X-ray

respectively.

Figure 1: Normal chest X-Ray.

Figure 2: Pneumonia chest X-Ray.

3.2 Working

The methodical process of the pneumonia detection

system begins with the dataset preparation, which

utilizing a labeled dataset of chest X-ray images. The

three subsets of the dataset are the training set,

validation set, and test sets. The test set provides an

accurate and fair evaluation of the model’s

performance, the validation set supports

hyperparameter optimization and performance

tracking, and the training set is used to train the

model. This prevents overfitting and provides an

effective training process. In the next stage, VGG16

is utilized as the fundamental model to develop the

architecture of the model. The custom dense layers

with ReLU activation are added to VGG16, a pre-

trained model of CNN because to improve non-

linearity and learning capacity and also dropout layers

are employed to improve model generalization by

reducing the risk of overfitting through the random

deactivation of neurons during the training period.

The final output layer predicts the chest X-ray image

which shows whether it is ‘Normal’ or ‘Pneumonia’

for binary classification by using the sigmoid

activation function. This model is been trained using

the augmented data of the image, that goes through

Pneumonia Detection Using Deep Learning

stability-improving changes such as rescaling,

flipping and rotation. By using the Adam optimizer,

the learning rates are continuously adjusted to

optimize the convergence speed. The early stopping

process is used to prevent overfitting and unwanted

computation by finishing the training process when

validation loss stabilizes. The continuous learning is

maintained by assessing the training process using

accuracy metrics and loss curves. Once the training

process is completed then the evaluation phase

utilizes a various metrics to evaluate the model's

performance, including accuracy, precision, recall,

and F1-score, which evaluates the model’s

effectiveness in data classification on the test dataset.

Additionally, the prediction accuracy is analyzed by

evaluating true positives, false positives, true

negatives, and false negatives using a confusion

matrix. During the deployment phase, the trained

model is integrated into a Flask-based web interface,

providing a real-time communication. The web-based

application enables users to upload the chest X-ray

images, which are then processed by the model to

predict the condition of pneumonia. This interface

provides an efficient automated diagnostic system

that helps healthcare professionals by making faster

and more accurate decisions.

4 RESULT AND ANALYSIS

Figure 3: Training & validation loss graph.

The graph in figure 3 shows the training and

validation loss during the training period. The blue

line which is represented as training loss gradually

decreases, showing that the model improves and

processes more data. The orange line which

represented as validation loss fluctuates in the

beginning but afterward, it becomes steady at a low

value. This shows that the model generalizes well to

new data. The early fluctuations in validation loss

indicate slight uncertainty, but there are no major

signs of overfitting. The consistently stable

downward trend confirms an effective learning.

Figure 4: Training & validation accuracy graph.

The graph in figure 4 which mentioned above

signifies the training and validation accuracy. The

training accuracy line increases easily and attains a

high value above 95%. On the other hand, the

validation accuracy line fluctuates in the beginning

before reaching close to 100%. These fluctuations

imply that the model in the early stages had difficulty

with validation data but eventually adjusted well. For

both the training and new data the final accuracy

levels indicate that the model is providing accurate

predictions, pointing to have a good generalization.

Figure 5: Confusion matrix.

The image in figure 5 shown above is the confusion

matrix, which helps us to understand that how

accurately the model classifies the chest X-ray

images. Among all the normal cases, 199 were

correctly detected, yet 35 cases were misclassified as

pneumonia. In the case of pneumonia, the model

correctly recognized 383, but 7 wrongly identified as

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

normal. The least number of false negatives, the 7

cases only is highly important because this indicates

that the model almost never misses actual pneumonia

cases, which is important for real-world medical

applications. On the other hand, the false positives

which are 35 cases that are higher, this is actually

better because it reduces the chance of missing

pneumonia cases.

Figure 6: Classification report & prediction result.

The above figure 6 is a classification report of the

pneumonia detection model that highlights the

model’s effective performance.

• The precision for the normal cases is 0.97,

which means the 97% of predicted normal cases

are exactly normal.

• The recall for the normal cases is 0.85, which

means the 85% of actual normal cases were

correctly identified.

• The precision for the pneumonia cases is 0.92,

and recall is 0.98, showing that the model

identifies pneumonia cases in a very effective

way.

• The model achieves an overall accuracy of 93%,

indicating high model performance.

• The macro and weighted averages of precision,

recall, and F1-score are approximately 0.93,

verifying equal performance for both classes.

At the bottom of the classification report, the note

‘Model saved successfully!’ shows that the trained

model has been saved and can be used later in future.

The last part that is the prediction output which says

the ‘Person is safe’ shows that the uploaded image of

chest X-ray was correctly classified as normal, with

the prediction value [[0.]] likely falls into the normal

category in the model’s processing system.

Figure 7: Web interface for pneumonia detection.

This above-mentioned figure 7 is a Flask-based web

interface which is simple and user-friendly to enable

users to upload the chest X-ray images for pneumonia

detection. This interface is running on local host

(127.0.0.1:5000) on your computer, this makes that it

is only accessible from your own machine during

testing locally. By uploading an image through the

file upload button, and then pressing the ‘Predict’

button, users can have the model identify whether it

is normal or pneumonia. After the prediction, the

application will provide a result such as ‘Person is

safe’ or ‘Person is has Pneumonia’, ensuring the

output is clear and easy to understand. This

deployment is an easy access to users, allowing

anyone to test the chest X-ray images easily without

having the direct interaction with the code, making it

an effective tool for quick and real-time diagnosis.

5 CONCLUSIONS

This research used CNN and VGG16 to build a deep

learning model for pneumonia detection system. This

model achieved high accuracy in classifying the chest

X-ray images, demonstrating the capability of deep

learning in the field of medical diagnosis. A web-

based interface developed to make the system utilized

in real-time experience, providing a faster and

effective automatic diagnostic system. The future

enhancements may include using an extensive dataset

for higher accuracy and faster model performance. In

summary, this study highlights AI-driven systems

that supports in medical field assisting doctors in

treating and detecting pneumonia at early stage.

REFERENCES

Asnake, Nigus Wereta, Ayodeji Olalekan Salau, and Aleka

Melese Ayalew. "X-ray image-based pneumonia

detection and classification using deep

learning." Multimedia Tools and Applications 83.21

(2024): 60789-60807.

Ayan, Enes, and Halil Murat Ünver. "Diagnosis of

pneumonia from chest X-ray images using deep

learning." 2019 Scientific meeting on electrical-

electronics & biomedical engineering and computer

science (EBBT). Ieee, 2019.

Barhoom, Alaa MA, and Samy S. Abu-Naser. "Diagnosis

of pneumonia using deep learning." (2022).

Gabruseva, Tatiana, Dmytro Poplavskiy, and Alexandr

Kalinin. "Deep learning for automatic pneumonia

detection." Proceedings of the IEEE/CVF conference

on computer vision and pattern recognition workshops.

2020.

Pneumonia Detection Using Deep Learning

Goyal, Shimpy, and Rajiv Singh. "Detection and

classification of lung diseases for pneumonia and

Covid-19 using machine and deep learning

techniques." Journal of Ambient Intelligence and

Humanized Computing 14.4 (2023): 3239-3259.

Jain, Deepak Kumar, et al. "Deep Learning‐Aided

Automated Pneumonia Detection and Classification

Using CXR Scans." Computational intelligence and

neuroscience 2022.1 (2022): 7474304.

Kundu, Rohit, et al. "Pneumonia detection in chest X-ray

images using an ensemble of deep learning

models." PloS one 16.9 (2021): e0256630.

Li, Shaojie, Yuhong Mo, and Zhenglin Li. "Automated

pneumonia detection in chest x-ray images using deep

learning model." Innovations in Applied Engineering

and Technology (2022): 1-6.

Manickam, Adhiyaman, et al. "Automated pneumonia

detection on chest X-ray images: A deep learning

approach with different optimizers and transfer learning

architectures." Measurement 184 (2021): 109953.

Pant, Ayush, et al. "Pneumonia detection: An efficient

approach using deep learning." 2020 11th International

Conference on Computing, Communication and

Networking Technologies (ICCCNT). IEEE, 2020.

Saul, Can Jozef, Deniz Yagmur Urey, and Can Doruk

Taktakoglu. "Early diagnosis of pneumonia with deep

learning." arXiv preprint arXiv:1904.00937 (2019).

Sharma, Shagun, and Kalpna Guleria. "A deep learning

based model for the detection of pneumonia from chest

X-ray images using VGG-16 and neural

networks." Procedia Computer Science 218 (2023):

357-366.

Singh, Swapnil. "Pneumonia detection using deep

learning." 2021 4th Biennial International Conference

on Nascent Technologies in Engineering (ICNTE).

IEEE, 2021.

Tilve, Ashitosh, et al. "Pneumonia detection using deep

learning approaches." 2020 international conference on

emerging trends in information technology and

engineering (ic-ETITE). IEEE, 2020.

Yaseliani, Mohammad, et al. "Pneumonia detection

proposing a hybrid deep convolutional neural network

based on two parallel visual geometry group

architectures and machine learning classifiers." IEEE

access 10 (2022): 62110-62128.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES