Identification of Lung Disease via X-Ray Images Using Knowledge Distillation and Vision Transformer

Lung disease is a widespread health concern affecting millions of people worldwide. Accurate diagnosis is essential for effective treatment, yet it often requires expert interpretation of chest X-rays to identify specific disease types. Unfortunately, many healthcare facilities, particularly in remote or resource-limited areas, lack access to skilled radiologists, leading to delayed diagnoses and potentially poor patient outcomes. To address this gap, this research introduces an automated lung disease classification system that leverages knowledge distillation for accurate and efficient diagnosis. The proposed system employs the Vision Transformer (ViT) architecture as the teacher model and MobileNet as the student model, aiming to classify chest X-ray images into five categories: normal, COVID-19, viral pneumonia, bacterial pneumonia, and tuberculosis. Knowledge distillation enables the student model to learn critical diagnostic features from the teacher model, resulting in a compact, high-performing model suitable for real-world deployment in settings with limited computational resources. Experimental evaluation on an original dataset of X-ray images demonstrated a classification accuracy of 93.07%, with high precision, recall, and F1-score, indicating robust performance across multiple disease types. This system holds promise as a valuable tool for assisting healthcare providers in making quick, accurate diagnoses, potentially improving treatment timelines and outcomes for patients in underserved regions.