Comparative Study of Deep Learning Techniques for Automated Classification of Lung Diseases

• Published in: 2023 4th International Conference on Smart Electronics and Communication (ICOSEC) pp. 1324 - 1328
• Main Authors: H, Rekha, Kumaravel, T., Natesan, P., B M, Brinda, Sangeetha, S., Dharanesh, S
• Format: Conference Proceeding
• Language: English
• Published: IEEE 20.09.2023
• Subjects: Data augmentation; Deep learning; Lung; Pre-processing; Pulmonary diseases; Real-time systems; Surveys; Transfer learning; Visual Geometry Group (VGG); Visualization; X-ray image Dataset
• DOI: 10.1109/ICOSEC58147.2023.10276053

Accurate classification of lung requires overcoming limited availability of labeled data, particularly for rare conditions like COVID-19, and the risk of overfitting due to the high dimensionality of medical images. The use of transfer learning with VGG16, VGG19, and DenseNet201 architectures enables leveraging pre-trained models' knowledge of general image features, allowing effective learning on a smaller medical image dataset and reducing the risk of overfitting. This project proposes three deep learning methods based on the VGG16, VGG19, and DenseNet201 architectures. These approaches aim to automatically classify these three lung diseases using chest X-ray images. To enhance the data, preprocessing techniques and image augmentation methods were employed. The VGG16, VGG19, and DenseNet201 models were trained using transfer learning on this dataset, and performance was evaluated on a separate test set. The achieved accuracies for VGG16, VGG19, and DenseNet201 models were 95%, 96%, and 97%, respectively. These impressive accuracies demonstrate the capacity of the approaches for precise and early detection of lung diseases. Among the three architectures, DenseNet201 consistently exhibited the highest performance for all three disease classifications. In summary, the proposed deep learning techniques utilizing VGG16, VGG19, and DenseNet201 architectures offer promising tools for the automated classification of lung diseases based on chest X-ray images. Their implementation can greatly assist medical professionals in promptly diagnosing and treating these diseases. DenseNet201, in particular, showcased superior performance among the three architectures.