December 17, 2025Frontiers in MedicineOpen Access

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

View Full Paper

Ask AI

Bookmark

Why the study?

Lung disease classification plays a significant part in the early discovery and determination of respiratory conditions.

Structured PICO

Population

10,425 X-ray images from the Lung X-Ray Image Dataset categorized into three classes: Normal (3,750 images), Lung Opacity (3,375 images), and Viral Pneumonia (3,300 images)

Intervention

MedViT and Swin Transformer deep learning models with geometric and photometric data augmentation and Kullback-Leibler Divergence optimization

Outcome

Classification accuracy

Transformer-based models, specifically MedViT with Kullback-Leibler Divergence optimization, demonstrate high accuracy for automated lung disease classification from X-ray images.

Implication

May aid automated lung opacity and pneumonia detection on X-rays; leaves open prospective clinical validation before adoption.

Abstract

Introduction: Lung disease classification plays a significant part in the early discovery and determination of respiratory conditions. Methods: This paper proposes a novel approach for lung disease classification utilizing two advanced deep learning models, MedViT and Swin Transformer, applied to the Lung X-Ray Image Dataset that includes 10,425 X-ray images categorized into three classes: Normal with 3,750 images, Lung Opacity with 3,375 images, and Viral Pneumonia with 3,300 images. A series of data augmentation methods, including geometric and photometric augmentation, are applied to improve model performance and generalization. Results: The results illustrate that both MedViT and Swin Transformer accomplish promising classification accuracy, with MedViT showing particular strength in medical image-specific feature learning due to its hybrid convolutional and transformer design. The impact of different loss functions is also examined, where Kullback-Leibler Divergence yields the highest accuracy and effectively handles class imbalance. The best-performing MedViT model achieves an accuracy of 98.6% with a loss of 0.09. Discussion: These findings highlight the potential of transformer-based models, particularly MedViT, for reliable clinical decision support in automated lung disease classification.