Lung cancer detection using transfer learning and hybrid deep learning models: A review

Introduction

Lung cancer is one of the leading causes of cancer-related mortality worldwide. According to global health reports, millions of new cases are diagnosed annually, and a significant proportion of patients are detected at advanced stages. Early diagnosis is often difficult because symptoms may remain unnoticed during the initial stages of the disease. Consequently, there is a strong need for automated diagnostic systems capable of identifying lung abnormalities at an early stage. Medical imaging has become a fundamental component of lung cancer diagnosis. CT scans provide detailed information about lung tissues and nodules, while PET scans offer metabolic information that helps distinguish malignant tissues from normal tissues. However, manual examination of thousands of image slices is labor-intensive and susceptible to observer variability. Artificial Intelligence has emerged as a powerful tool in healthcare. Deep learning algorithms can automatically learn complex image features and classify medical images with high accuracy. Transfer learning allows pretrained networks to be adapted to medical imaging tasks, reducing training time and data requirements. Hybrid deep learning systems combine multiple architectures to improve robustness and diagnostic performance. This review investigates recent developments in lung cancer detection using transfer learning and hybrid deep learning models. The primary objective is to analyze current methodologies, compare model performances, identify existing limitations, and discuss future opportunities for research and clinical deployment.

Motivation of the Study

The motivation for this review arises from the increasing prevalence of lung cancer and the growing need for reliable automated diagnostic systems. Although numerous studies report high classification accuracy, challenges remain regarding generalization, Explain ability, clinical validation, and deployment in real-world environments. Many existing systems are evaluated using limited datasets, and their performance may vary across different imaging centers. Furthermore, clinicians require interpretable AI models that provide explanations for diagnostic decisions. Therefore, understanding current research trends and identifying future directions are essential for developing clinically useful AI systems.

Objective: The primary objective of this review is to examine recent advancements in transfer learning and hybrid deep learning approaches for lung cancer detection. It aims to compare various artificial intelligence models, including Convolutional Neural Network (CNN)-based, transfer learning, ensemble, and hybrid architectures, to evaluate their effectiveness in improving diagnostic accuracy. Furthermore, the study seeks to analyze the datasets, preprocessing methods, and image segmentation techniques commonly employed in lung cancer detection research. Another important objective is to identify the existing research gaps and limitations in current studies, highlighting challenges related to model performance, generalization, and clinical applicability. Additionally, the review discusses potential future directions for AI-assisted lung cancer diagnosis and proposes a review-based framework to guide future research toward the development of more accurate, efficient, and reliable diagnostic systems.

Background

A. Lung Cancer and Medical Imaging

Lung cancer occurs due to uncontrolled growth of abnormal cells within lung tissues. It is generally classified into Small Cell Lung Cancer (SCLC) and Non-Small Cell Lung Cancer (NSCLC). CT imaging is the most commonly used modality for lung cancer screening because it provides detailed crosssectional views of lung structures. Medical imaging enables doctors to examine internal lung structures without the need for invasive procedures. CT scans are widely preferred because they provide detailed crosssectional images of the lungs. Other imaging modalities, such as PET scans, offer metabolic information that can help identify cancer activity, while MRI and chest X-ray imaging can support diagnosis and treatment planning in specific clinical situations. Each imaging technique has its own advantages and limitations in terms of image quality, resolution, contrast, and diagnostic capability. Therefore, artificial intelligence-based lung cancer detection systems should be designed according to the characteristics of the imaging modality being used. Factors such as image resolution, contrast, noise levels, and the specific diagnostic objective play an important role in developing accurate and reliable AI models for medical image analysis.

B. Deep Learning in Medical Imaging:

Deep learning has revolutionized medical image analysis. CNNs automatically extract hierarchical image features without manual feature engineering. Early layers capture low-level patterns such as edges and textures, while deeper layers identify complex anatomical structures and disease-specific features. In lung cancer detection, CNN models can learn important visual characteristics such as edges, textures, shapes, lesion boundaries, and abnormal tissue patterns from medical images. The initial layers of a CNN typically capture lowlevel features, including edges and simple textures, while deeper layers learn more complex and high-level structures associated with disease characteristics. This hierarchical feature learning enables CNNs to identify subtle visual differences that may not be easily detectable by traditional methods. The ability of CNNs to automatically learn discriminative features makes them highly effective for medical image classification tasks. As a result, deep learning-based CNN models have become a popular approach for lung cancer detection, offering improved diagnostic accuracy and reducing the need for manual feature extraction.

C. Data Augmentation in Lung Cancer Detection

Data augmentation is widely used to improve model generalization. Since medical image datasets are often limited, augmentation creates modified versions of existing images without changing their medical significance. Common augmentation techniques include rotation, flipping, zooming, cropping, shifting, brightness adjustment, contrast enhancement, and noise injection. Advanced augmentation methods such as Random Pixel Swap, mixup, cutout, and generative augmentation have also been explored in recent studies. These techniques improve model robustness by exposing the network to a wider range of image variations and reducing the risk of over fitting. Data augmentation can also help address class imbalance by increasing the diversity of underrepresented classes in the dataset. However, augmentation must be applied carefully in medical imaging because unrealistic image transformations may distort clinically important features and lead to inaccurate predictions. Properly designed augmentation strategies preserve diagnostic information while enhancing the model's ability to generalize to unseen clinical data. Separates lung regions or nodules from the background.

D. Transfer Learning

General Workflow of Lung Cancer Detection

References

1. Shatnawi MQ, Abuein Q, Al-Quraan R. Deep learningbased approach to diagnose lung cancer using CT-scan images. Intelligence-Based Medicine. 2025;11:100188.
2. Simie E, Kaur M. Lung cancer detection using convolutional neural network (CNN). International Journal of Advance Research, Ideas and Innovations in Technology. 2019;5(4).
3. Shah AAM, Malik HAM, Muhammad A, Alourani A, Butt ZA. Deep learning ensemble 2D CNN approach towards the detection of lung cancer. Scientific Reports. 2023;13(1):2987.
4. Sait AR. Lung cancer detection model using deep learning technique. Applied Sciences. 2023;13(22):12510.
5. Thaseen M, et al. A review of using CNN approach for lung cancer detection through machine learning. In: Proceedings of the 2nd International Conference on Advance Computing and Innovative Technologies in Engineering; 2022. Piscataway (NJ): IEEE; 2022.
6. Takale DG, et al. Lung cancer detection by using CNN architecture models. In: Proceedings of the International Conference on Artificial-Business Analytics, Quantum and Machine Learning. Singapore: Springer Nature Singapore; 2023.
7. Abe AA, Nyathi M. Lung cancer diagnosis from computed tomography images using deep learning algorithms with random pixel swap data augmentation: Algorithm development and validation study. JMIR Bioinformatics and Biotechnology. 2025;6(1):e68848.
8. Faruqui N, Yousuf MA, Whaiduzzaman M, Azad AKM, Barros A, Moni MA. LungNet: A hybrid deep-CNN model for lung cancer diagnosis using CT and wearable sensor-based medical IoT data. Computers in Biology and Medicine. 2021;139:104961.
9. Liz-López H, de Sojo-Hernández ÁA, D'AntonioMaceiras S, Díaz-Martínez MA, Camacho D. Deep learning innovations in the detection of lung cancer: Advances, trends, and open challenges. Cognitive Computation. 2025;17(2):67.
10. Klangbunrueang R, Pookduang P, Chansanam W, Lunrasri T. AI-powered lung cancer detection: Assessing VGG16 and CNN architectures for CT scan image classification. In: Informatics. 2025;12(1):18.
11. Nimmagadda SM, Suryanarayana G, Kumar PV, Vamsi GS, Imla DA, Harika C. Early detection of lung cancer using deep learning techniques: A comprehensive review. Archives of Computational Methods in Engineering. 2025:1-32.
12. Chakrabarty N, Mahajan A, Shetty N, Mummudi N, Niyogi D, Hota F, et al. Imaging patterns and recommendations for diagnosis, staging, and management of lung cancer. BJR Open. 2025;7(1):tzaf013.
13. Abdullahi K, Ramakrishnan K, Ali AB. Deep learning techniques for lung cancer diagnosis with computed tomography imaging: A systematic review for detection, segmentation, and classification. Information. 2025;16(6):451.
14. Muñoz-Aseguinolaza U, Fernandez-Iriondo I, RodríguezMoreno I, Aginako N, Sierra B. Convolutional neural network-based classification and monitoring models for lung cancer detection: 3D perspective approach. Heliyon. 2023;9(11).