Deep Feature Learning for Lung Disease and Lung Patches Detection from Chest CT and X-Ray Scans Using GANs

Abstract

Accurate and efficient classification of lung diseases from chest CT and X-ray images is critical for timely diagnosis and treatment. This study proposes a novel hybrid framework that combines unsupervised representation learning via Generative Adversarial Networks (GANs) with supervised feature extraction using a pre-trained VGG16 convolutional neural network. The GAN component learns latent structural features from large amounts of unlabeled medical images, while VGG16 extracts high-level semantic features from labeled data. By fusing these complementary feature representations, the system achieves improved classification accuracy across multiple lung disease categories, including COVID-19, pneumonia, tuberculosis, and lung cancer. Experimental evaluations demonstrate the effectiveness of the proposed approach in leveraging both labeled and unlabeled data, reducing dependency on extensive manual annotations. The framework shows promising potential for scalable, automated lung disease diagnosis in clinical settings.

Introduction

1. Global Context

• Lung diseases are a major global health issue, responsible for over 7.1 million deaths annually, ranking as the fourth leading cause of death globally (WHO).

• Causes include smoking, air pollution, infections, genetics, and autoimmune conditions.

• Common lung diseases: Asthma, Pneumonia, Tuberculosis, COPD, Lung Cancer, and COVID-19.

• Early diagnosis and treatment are critical for better patient outcomes.

2. Diagnostic Tools

• Medical imaging modalities such as Chest X-rays (CXR), CT scans, MRI, and ultrasound are essential for diagnosing lung diseases.

• CT scans offer high-resolution, 3D imaging, useful for identifying nodules, inflammation, or structural damage.

• X-rays are commonly used for quick, cost-effective screening, despite small radiation exposure risks.

3. Deep Learning in Lung Disease Classification

• Manual analysis of medical images is subjective, slow, and can suffer from human error.

• Deep learning (DL), especially Convolutional Neural Networks (CNNs), are effective for automatic image-based diagnosis.

• However, supervised DL models require large amounts of labeled data, which are hard to obtain in medical domains.

4. Proposed Solution: Lung-GANs

A novel hybrid deep learning framework combining:

• Unsupervised learning (GANs): Learns lung image features without labels.

• Supervised learning (VGG16 CNN): Extracts high-level semantic features.

• Feature fusion: Combines GAN and VGG16 features to create a robust representation.

• Classification: Uses classifiers like SVM to detect lung diseases accurately.

Advantages:

• Reduces dependency on large labeled datasets.

• Captures both structural (GAN) and semantic (VGG16) features.

• Improves classification accuracy and robustness.

5. Methodology Overview

1. Data Preprocessing: Resize, normalize, augment images.

2. GAN Training: Generator creates synthetic lung images; Discriminator learns key lung features.

3. VGG16 Feature Extraction: Pre-trained CNN extracts semantic features from medical images.

4. Feature Fusion: Combine GAN and VGG16 outputs.

5. Classification: SVM or neural network predicts disease class.

6. Explainability: Techniques like Grad-CAM visualize areas important for predictions.

6. Results

• The Lung-GANs system was tested on six public datasets and outperformed other unsupervised models.

• Achieved 100% classification accuracy, demonstrating its effectiveness and scalability.

• Supports rapid clinical decisions, especially useful during outbreaks (e.g., COVID-19).

7. Literature Review Highlights

Conclusion

In this study, we proposed a hybrid framework that leverages unsupervised feature learning through Generative Adversarial Networks (GANs) alongside the supervised deep feature extraction of VGG16 for effective lung disease classification from chest CT and X-ray images. The integration of GAN-derived latent features with VGG16’s semantic representations allow the system to learn robust and comprehensive image features, improving classification accuracy while reducing reliance on extensive labeled datasets. This approach demonstrated promising results across multiple lung disease categories, highlighting its potential for clinical application in automated diagnosis. For future work, we aim to extend the framework by incorporating transformer-based architectures, such as Vision Transformers (ViTs), to capture long-range dependencies and further enhance feature representation. Additionally, exploring advanced GAN variants for better image synthesis and domain adaptation will be considered to improve generalization. Another important direction is to integrate explainability techniques more deeply, providing interpretable visualizations that can assist clinicians in understanding model decisions. Finally, expanding the dataset to include a wider variety of lung diseases and multi-modal imaging data will help build a more versatile and robust diagnostic tool.

References

[1] N. Dey, Y.-D. Zhang, V. Rajinikanth, R. Pugalenthi, and N. S. M. Raja, “Customized VGG19 architecture for pneumonia detection in chest X-rays,” Pattern Recognit. Lett., vol. 143, pp. 67–74, 2021. [2] X. Mei, “Artificial intelligence-enabled rapid diagnosis of patients with Covid-19,” Nature Med., vol. 26, no. 8, pp. 1224–1228, 2020. [3] D. Ezzat, A. E. Hassanien, and H. A. Ella, “An optimized deep learning architecture for the diagnosis of COVID-19 disease based on gravitational search optimization,” Appl. Soft Comput., vol. 98, p. 106742, 2020. [4] .R. Zhang, “COVID-19XrayNet: A two-step transfer learning model for the COVID-19 detecting problem based on a limited number of chest X-ray images,” Interdiscipl. Sci.: Comput. Life Sci., vol. 12, no. 4, pp. 555–565, 2020. [5] I. D. Apostolopoulos, S. I. Aznaouridis, and M. A. Tzani, “Extracting possibly representative COVID-19 biomarkers from X-ray images with a deep learning approach and image data related to pulmonary diseases,” J. Med. Biol. Eng., vol. 40, pp. 462–469, 2020. [6] Aviles-Rivero, P. Sellars, C.-B. Schönlieb, and N. Papadakis, “Graphxcovid: Explainable deep graph diffusion pseudo-labelling for identifying COVID-19 on chest X-rays,” 2021, arXiv preprint arXiv:2010.00378v2. [7] Richa Sharma, Monika Mangl, Priyanca Gonsalves,Lung Disease Detection from Chest X-Ray using GANs.IEEE Xplore Part Number: CFP24CV1-ART; ISBN: 979-8-3503-2753-32025. [8] M. Ahmad et al., “Comparative Study of Vision Transformers for Chest X-Ray Classification,” arXiv preprint arXiv:2503.18973, 2025. [9] Yan Kuang 1,2, Tian Lan 1,3, (Member, Ieee), Xueqiao Peng 1 , Gati Elvis Selasi 1, Qiao Liu 1, (Member, Ieee), And Junyi Zhang 2 [10] S. Unsupervised Multi-Discriminator Generative Adversarial Network for Lung Nodule Malignancy ClassificationMay 8, 2020. [10] Pooja Yadav; Neeraj Menon; Vinayakumar Ravi; Sowmya Vishvanathan .Lung-GANs: Unsupervised Representation Learning for Lung Disease Classification Using Chest CT and X-Ray Images.Publisher: IEEE. [11] Salma Sultana; Anik Pramanik; Md. Sadekur Rahman.Lung Disease Classification Using Deep Learning Models from Chest X-ray ImagesPublisher: IEEE [12] H. Greenspan, B. van Ginneken, and R. M. Summers, “Guest Editorial Deep Learning in Medical Imaging: Overview and Future Promise of an Exciting New Technique,” IEEE Trans. Med. Imaging, vol. 35, no. 5, pp. 1153–1159 [13] Hui Ding, Lingyan Fan, Jingfeng Zhang, jingfengzhang73@163.com, and Guosheng Gao. Deep Learning-Based System Combining Chest X-Ray andComputerizedTomography Images for COVID-19 Diagnosis. British Journal of Hospital MedicineVolume 85, Number 8https://doi.org/10.12968/hmed.2024.0244. [14] Real-Time Lung Cancer Detection Using GAN-Enhanced LDCT Imaging and CNN-Based Classification for Clinical Accuracy.Anupriya K; Ramyakumari T; Saisakthi B; Kavyadharshini M2024 International Conference on System, Computation, Automation and Networking (ICSCAN)

Copyright

Copyright © 2025 Jibi P John, T Venkatesh. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.