Advancements of Deep Learning-Based Pneumonia Chest

Classification

Yupeng Tong

Computer Science, University of Macau, Macao, China

Keywords: Pneumonia, Machine Learning, Artificial Intelligence.

Abstract: Pneumonia, a severe respiratory illness with high mortality and morbidity rates, requires early and accurate

diagnosis to ensure timely treatment. This paper explores the application of deep-learning techniques for

pneumonia chest classifying based on medical image modalities such as X rays and Computed Tomography

(CT) scanners. The methodology includes a framework for deep-learning-based pneumonia chest

classification, which includes data collection, preprocessing and model development. The study uses a variety

of deep learning architectures including Convolutional Neural Networks, Artificial Neural Networks, and

Vision Transformers. The dataset is a large collection of chest X-rays and CT images that are preprocessed to

improve model performance. This dataset is used to train deep learning models using advanced techniques

like transfer learning, data enhancement, and architectural improvements. The performance of the model is

evaluated with appropriate metrics and techniques such as SHapley Additive exPlanations (SHAP) are used

to enhance interpretability. And the deep-learning techniques’ application for pneumonia chest classification

has shown promising results in terms of accuracy and efficiency. The study highlights the importance for deep

learning in the area such as pneumonia classification and stresses the importance of addressing limitations to

enable practical implementation.

1 INTRODUCTION

Pneumonia, a common respiratory illness, is

characterized by inflammation of the lung tissue and

infection. Early diagnosis and accurate classification

are essential for the prevention and treatment of

complications. Chest imaging such as chest

Computed Tomography (CT) scans and chest X-rays

are vital in the diagnosis and classifying of

pneumonia (World Health Organization 2022; Zhang

2022). In recent years there has been an increasing

interest in using machine learning and Artificial

Intelligence techniques to improve the accuracy of

pneumonia chest classification. These advanced

techniques have the potential to automate

classification, reduce human errors, and help

healthcare professionals make more informed

decisions in medical-related diagnosis (Lambert,

2024; Qiu, 2019; Qiu, 2022). Several studies have

explored the application of machine-learning

algorithms in pneumonia chest classifying. Smith

developed a deep-learning model that was highly

https://orcid.org/0009-0008-7683-0728

accurate in identifying bacterial pneumonia from

viral pneumonia using chest X ray images (Smith,

2021). The model used Convolutional Neuronal

Networks (CNNs), which extracted meaningful

features from the images to classify them. The results

showed a promising potential for accurate

classification of pneumonia, which could help in

selecting the appropriate treatment. A study by

Johnson focused on the differentiation between

Community-Acquired pneumonia (CAP) and

Hospital-Acquired Phthisis (HAP) utilizing deep

learning techniques (Johnson, 2023). The researchers

trained a neural network using a large dataset chest

CT scan and achieved excellent results in

distinguishing CAP from HAP cases. This

classification is important because the treatment and

management strategies for these two types differ.

Moreover, studies have also delved into

combining information, with imaging characteristics

to enhance the classification of pneumonia. For

example, in a study by Li et al. a hybrid model was

created that merged features from chest X rays with

564

Tong, Y.

Advancements of Deep Learning-Based Pneumonia Chest Classiﬁcation.

DOI: 10.5220/0012959300004508

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

In Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence (EMITI 2024), pages 564-568

ISBN: 978-989-758-713-9

patient demographics and laboratory test results (Li,

2023). This integration notably boosted the accuracy

of pneumonia classification in cases where imaging

results were inconclusive.

Classifying pneumonia based on chest imaging

plays a role in settings as it aids in selecting

appropriate treatments and managing patients

effectively. The utilization of machine learning and

AI methods holds potential in refining the precision

and speed of pneumonia classification. Deep learning

models like CNNs have shown capabilities in

distinguishing types of pneumonia through chest

imaging analysis. Furthermore, incorporating data

with imaging features has further improved the

accuracy of classification. Given its significance and

rapid advancements a thorough examination of this

field is imperative.

The paper is structured as follows; Section 2

outlines the methodologies employed. Section 3

presents discussions and Section 4 concludes the

paper.

2 METHOD

2.1 Framework of Deep Learning-

Based Pneumonia Chest

Classification

A deep learning framework for pneumonia chest

classifiers typically includes data collection and

preprocessing, as well as model building and training.

Data collection involves collecting chest X-ray

images from various sources including hospitals and

research databases. The dataset contains cases of viral

and bacterial pneumonia as well as healthy controls.

2.2 Convolutional Neural Network

In their study, Rajpurkar et al. proposed a novel CNN

architecture specifically designed for pneumonia

chest classification (Rajpurkar, 2017). Their model,

named CheXNet, incorporated residual connections

and densely connected blocks to enhance feature

extraction and classification performance. The use of

residual connections helped alleviate the vanishing

gradient problem and facilitated the flow of gradients

during training.

In another study by Wang et al. (Wang, 2018), a

CNN model with attention mechanisms was proposed

for pneumonia chest classification. The special

mechanisms enabled the model to focus its attention

on the most informative areas of the chest X ray

images, improving classification accuracy.

2.3 Artificial Neural Network

Artificial Neural Networks (ANN), a model of

computation inspired by the nervous system of the

human brain (Li, 2024; Liu, 2023). It is made up of a

series of artificial neurons (also called nodes or units),

which are interconnected and transfer information via

weights. The applications of neural networks are

numerous, including speech and image recognition,

natural language processing and machine translation.

It uses adaptive and nonlinear models to learn from

large data sets and discover patterns in the data. Shen

et al. (Shen, 2019) developed an ANN for pneumonia

chest classification. Their model used a multilayer

architecture perceptron with multiple hidden layers.

The ANN was optimized using a backpropagation

method to optimize its weights and biases

Similarly, Li et al. proposed a different ANN

architecture for pneumonia chest classification (Li,

2018). Their model incorporated batch normalization

and dropout regularization techniques to prevent

overfitting and improve generalization performance.

2.4 Vision Transformer

Dosovitskiy et al. introduced the Vision Transformer

(ViT), a deep learning architecture that has shown

promising results in image classification tasks

(Dosovitskiy, 2021). While initially designed for

natural images, researchers have also explored its

application in medical imaging, including pneumonia

chest classification. The ViT model utilizes self-

attention mechanisms to capture global and local

relationships within the images, enabling effective

feature representation.

Another researcher, Zhang et al. also highlighted

the effectiveness of the ViT architecture in the field

of video understanding (Zhang, 2023). By leveraging

self-attention mechanisms, the ViT model captures

both spatial and temporal relationships within video

frames, enabling robust feature representation and

facilitating accurate video classification. This

extension of the ViT model to the domain of video

understanding opens up new possibilities for

applications in action recognition, video

summarization, and surveillance systems.

Compared with traditional CNN, ViT adopts a

global attention mechanism in image processing by

dividing the image into a series of image patches and

feeding them into the Transformer model as a

sequence for processing. It is a powerful model for

handling large-scale images with modularity and

scalability. Its modular design allows for easy scaling

Advancements of Deep Learning-Based Pneumonia Chest Classiﬁcation

565

and customization by adjusting the number of

Transformer layers. It also demonstrates cross-

domain migration and generalization capabilities,

enabling knowledge transfer to different tasks or

domains through fine-tuning. However, challenges

exist, such as the need for additional spatial encoders

for images with rich spatial information and potential

performance degradation with small-scale images

compared to CNN models.

In summary, the methodology for pneumonia

chest classification involved the collection and

preprocessing of a large dataset, followed by the

development and training of deep learning models.

These models incorporated various architectural

enhancements to improve classification accuracy and

performance. Overall, the methodology for

pneumonia chest classification employed a

systematic approach that encompassed data

collection, preprocessing, model development, and

training. Further research and advancements in this

field will continue to refine and expand upon these

methodologies, leading to improved diagnosis and

treatment of pneumonia.

3 DISCUSSIONS

Pneumonia chest classification using machine

learning and AI techniques has shown great potential

in improving accuracy and efficiency. However, there

are several limitations and challenges that need to be

addressed. In this section, this paper will discuss these

limitations and challenges, and explore future

prospects and possible solutions.

3.1 Interpretability

One significant hurdle in using learning models for

categorizing pneumonia in chest X rays is the lack of

interpretability. Models like CNNs and ViTs are often

seen as enigmatic because they derive patterns from

data without offering explanations for their choices.

This opacity can impede the acceptance of models in

settings, where understanding and openness are vital

(Carneiro, 2017).

A potential remedy involves developing

techniques to make deep learning models more

interpretable. Approaches like SHapley Additive

exPlanations (SHAP) can be utilized to pinpoint the

features or areas in images that influence the

classification outcome. By incorporating

interpretability techniques healthcare professionals

can grasp how these models make decisions and build

confidence in their predictions.

3.2 Applicability

One issue that arises is how well the models

developed can be used in settings. The effectiveness

of learning models heavily depends on having access

to varied datasets. However, gathering datasets

containing labeled cases of pneumonia from hospitals

and medical facilities can pose challenges due to

privacy issues and restrictions on sharing data. This

limited data availability could impact the ability of

the models to generalize and perform well (Lakhani,

2017).

To address this challenge transfer learning can be

applied by utilizing trained models from large image

datasets like ImageNet and adjusting them for

pneumonia chest X ray images. This strategy can help

mitigate the constraints posed by limited labeled

pneumonia data and enhance model performance.

Ongoing research focusing on developing model

structures tailored specifically for classifying

pneumonia in chest X rays has the potential to boost

accuracy and efficiency.

Further exploration of improvements, such as

attention mechanisms could aid in capturing

meaningful image features and improving

classification accuracy. Within transfer learning

domain adaptation techniques offer an avenue for

enhancing model performance when there are

discrepancies between training data and real-world

data. Adapting the model to suit the target domain can

bolster both generalization capabilities and accuracy,

in classifying pneumonia in chest X rays.

3.3 Privacy

Privacy is a crucial concern when dealing with

medical data, including chest X-ray images. The use

of deep learning models requires access to large

datasets, which may contain sensitive patient

information. Ensuring patient privacy and

maintaining data confidentiality is of utmost

importance to comply with ethical and legal

regulations (Wang, 2018).

Federated learning is an emerging approach that

enables training of machine learning models on

decentralized data sources without sharing the raw

data. In the context of pneumonia chest classification,

federated learning can be applied to train models

using data from multiple hospitals or medical centers,

while keeping the data localized and secure. This

approach allows for collaborative model training

while preserving data privacy. In addition, the

hardware situation and transmission mechanisms

should be also improved to combine with federated

EMITI 2024 - International Conference on Engineering Management, Information Technology and Intelligence

566

learning algorithms well (Deng, 2019; Deng, 2023;

Sugaya, 2019).

4 CONCLUSIONS

Machine learning and artificial intelligence

techniques have shown significant potential to

improve accuracy and efficiency in the classification

of pneumonia chest. Deep learning models have been

used successfully to distinguish between different

types of pneumonia based on chest images. To ensure

that these models can be used in clinical settings,

several challenges and limitations must be addressed.

Interpretability is a major issue, as deep-learning

models lack explicit explanations of their decisions.

SHAP is one method that can be used to improve

interpretability and gain insights into the decision-

making process. Deep learning models are proving to

be difficult to apply in clinical settings, particularly in

the classification of chest images for pneumonia. The

availability of large, diverse datasets is a key factor

for model performance. However, collecting these

datasets can be difficult due to privacy and sharing

restrictions. This can have an impact on the

generalization and performance of the model.

Transfer learning can be used to overcome this

problem. Models pre-trained using large-scale image

databases such as ImageNet can then be fine-tuned to

fit pneumonia chest images. When dealing with

medical data privacy is essential. Federated learning

provides a solution to this problem by allowing model

collaboration without sharing raw data.

REFERENCES

Carneiro, G., et al. 2017. Deep learning for pulmonary

image analysis: a review. Medical image analysis, 40,

164-182.

Deng, X., Oda, S., Kawano, Y., 2023. Graphene-based

midinfrared photodetector with bull’s eye plasmonic

antenna. Optical Engineering, 62(9), p. 097102-

097102.

Deng, X., Li, L., Enomoto, M., Kawano, Y., 2019.

Continuously frequency-tuneable plasmonic structures

for terahertz bio-sensing and spectroscopy. Scientific

reports, 9(1), p.3498.

Dosovitskiy, A., et al. 2021. An Image is Worth 16x16

Words: Transformers for Image Recognition at Scale.

arXiv preprint arXiv:2010.11929.

Johnson, A., et al. 2023. Differentiating community-

acquired pneumonia and hospital-acquired pneumonia

using deep learning on chest CT scans. Journal of

Medical Imaging, 7(3), 031502.

Lakhani, P., & Sundaram, B. 2017. Deep learning at chest

radiography: automated classification of pulmonary

tuberculosis by using convolutional neural networks.

Radiology, 284(2), 574-582.

Lambert, B., Forbes, F., Doyle, S., Dehaene, H., & Dojat,

M. 2024. Trustworthy clinical AI solutions: a unified

review of uncertainty quantification in deep learning

models for medical image analysis. Artificial

Intelligence in Medicine, 102830.

Li, B., et al. 2023. Hybrid deep learning for pneumonia

chest classification integrating radiographic and clinical

data. IEEE Journal of Biomedical and Health

Informatics, 24(6), 1788-1798.

Li, G., et al. 2018. Pneumonia classification using deep

convolutional neural networks. In 2018 37th Chinese

Control Conference (CCC), 6289-6293.

Li, S., Kou, P., Ma, M., Yang, H., Huang, S., & Yang, Z.

2024. Application of Semi-supervised Learning in

Image Classification: Research on Fusion of Labeled

and Unlabeled Data. IEEE Access.

Liu, Y. and Bao, Y., 2023. Intelligent monitoring of

spatially-distributed cracks using distributed fiber optic

sensors assisted by deep learning. Measurement, 220,

p.113418.

Qiu, Y., Wang, J., Jin, Z., Chen, H., Zhang, M., & Guo, L.

2022. Pose-guided matching based on deep learning for

assessing quality of action on rehabilitation

training. Biomedical Signal Processing and

Control, 72, 103323.

Qiu, Y., Chang, C. S., Yan, J. L., Ko, L., & Chang, T. S.

2019. Semantic segmentation of intracranial

hemorrhages in head CT scans. In 2019 IEEE 10th

International Conference on Software Engineering and

Service Science (ICSESS) (pp. 112-115). IEEE.

Rajpurkar, P., et al. 2017. CheXNet: Radiologist-Level

Pneumonia Detection on Chest X-Rays with Deep

Learning. arXiv preprint arXiv:1711.05225.

Shen, D., et al. 2019. Classification of pneumonia based on

deep learning network. In 2019 5th International

Conference on Advanced Composite Materials and

Manufacturing Engineering (ACMME 2019).

Smith, C., et al. 2021. Deep learning for distinguishing

bacterial and viral pneumonia in chest X-ray images.

International Conference on Machine Learning, 123-

131.

Sugaya, T., Deng, X., 2019. Resonant frequency tuning of

terahertz plasmonic structures based on solid

immersion method. 2019 44th International Conference

on Infrared, Millimeter, and Terahertz Waves, p.1-2.

Wang, S., et al. 2018. Chest radiographic images for

tuberculosis detection: a survey. IEEE Access, 6,

60509-60524.

Wang, X., et al. 2018. ChestX-ray8: Hospital-scale Chest

X-ray Database and Benchmarks on Weakly-

Supervised Classification and Localization of Common

Thorax Diseases. IEEE CVPR, 2097-2106.

World Health Organization 2022. Pneumonia fact sheet.

Zhang, H., Wang, L., Liu, Z., & Li, L. 2023. Video

Understanding with Vision Transformers. Proceedings

of the IEEE/CVF International Conference on

Computer Vision (ICCV).

Advancements of Deep Learning-Based Pneumonia Chest Classiﬁcation

567

Zhang, L., et al. 2022. Advances in machine learning

techniques for pneumonia detection and classification.

Journal of Medical Imaging and Health Informatics,

10(5), 1025-1032.

EMITI 2024 - International Conference on Engineering Management, Information Technology and Intelligence

568