X-MedVIT: Interpretable Vision Transformers for Reliable Pneumonia Detection in Medical Imaging

Abstract— AI has become an important tool in medical diagnostic systems. Deep learning models often lack transparency that makes it difficult for clinics to use them. To address this, we propose X-MedVIT, an interpretablemedical imaging framework based on Vision Transformers (ViT). This system analyzes the chest X-ray images and issues a diagnosis and an explanation of the rationale behind the decision. It incorporates attention rollout, gradient times input saliency, and token-level relevance propagation. Unlike standard ViT models that do not take into account the generalisation performance, X - MedVIT also provides visual and content-related evidence that suggests which areas influenced the prediction. Our experimental public datasets we process shows an average accuracy of approximately 95%, demonstrating the reliable performance and boosting clinical trust and reasoning transparency. By offering predictions The system allows for digital health care solutions and helps to support global targets such as the United Nation Sustainable Development Goal, in order to improve health and access to diagnostics. Keywords— Vision Transformers, Explainable AI, Medical Diagnosis, Chest X-ray Analysis, Clinical Transparency.