Harnessing Deep Learning Techniques for Lung Disease Diagnosis: A Comparative Study of CNN, Transformer and Mamba Models

Pneumonia, an infectious lung condition caused by bacteria, viruses, or other microorganisms, significantly impacts both pediatric and geriatric populations. The COVID-19 pandemic has underscored the necessity for swift and accurate diagnostic tools, with over 760 million infections reported. This study investigates the role of artificial intelligence in lung X-ray image classification by comparing the performance of three Convolutional Neural Networks, two Transformer models, and the Vision Mamba model using a standardized dataset. The study’s objectives were to (1) identify the most effective models, (2) examine the impact of initializing models using transfer learning, and (3) evaluate the Vision Mamba model’s performance relative to conventional models. Using a Kaggle dataset containing four X-ray image categories—COVID-19, normal, lung opacity, and viral pneumonia—six models were trained and tested with and without transfer learning. Results indicated that the Swin Transformer model outperformed others, and transfer learning significantly enhanced model accuracy. Although the Vision Mamba model exhibited lower accuracy, its computational efficiency highlights its potential utility. This study provides critical insights into artificial intelligence applications in lung disease diagnosis, supporting clinicians with accurate diagnostic tools.