Secure pulmonary diagnosis using transformer-based approach to X-ray classification with KL divergence optimization.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Respiratory Medicine

Background:

• Accurate lung disease classification is crucial for early detection and treatment of respiratory conditions.
• Deep learning models offer potential for automating the analysis of medical images like X-rays.

Purpose of the Study:

• To evaluate the efficacy of MedViT and Swin Transformer models for classifying lung diseases from X-ray images.
• To compare the performance of these deep learning models on a large dataset of chest X-rays.
• To investigate the impact of data augmentation and loss functions on classification accuracy.

Main Methods:

• Utilized the Lung X-Ray Image Dataset with 10,425 images across Normal, Lung Opacity, and Viral Pneumonia categories.
• Applied advanced deep learning models: MedViT (hybrid convolutional and transformer) and Swin Transformer.
• Implemented data augmentation techniques (geometric and photometric) and analyzed the effect of different loss functions, including Kullback-Leibler Divergence.

Main Results:

• Both MedViT and Swin Transformer demonstrated high classification accuracy.
• MedViT exhibited superior performance in learning medical image-specific features.
• The best MedViT model achieved an accuracy of 98.6% with Kullback-Leibler Divergence loss, effectively managing class imbalance.

Conclusions:

• Transformer-based models, especially MedViT, show significant potential for automated lung disease classification.
• These models can serve as valuable tools for clinical decision support in diagnosing respiratory conditions.
• The findings support the integration of advanced AI in medical image analysis for improved healthcare outcomes.