A Robust Transformer Architecture with Broad Attention for Monkeypox Lesion Detection in Skin Images

Monkeypox (Mpox) is a viral zoonotic disease whose re-emergence led the World Health Organization (WHO) to declare a global health emergency, with over 89,000 confirmed cases reported worldwide between January 2022 and August 2023. The disease manifests with skin lesions that often resemble other dermatological conditions such as chickenpox, making early and accurate diagnosis challenging. Although PCR-based diagnostic methods are reliable, they require physical sampling and are not always feasible in all clinical environments. Advances in deep learning and medical image analysis offer promising alternatives for automated diagnosis. This study aims to develop a robust and efficient deep learning model for accurate monkeypox lesion detection using skin images, addressing the limitations of conventional CNNs and standard Vision Transformer (ViT) models. We propose a Robust Transformer Architecture with Broad Attention for monkeypox lesion detection. The model integrates vision transformers with an enhanced broad attention mechanism capable of capturing both fine-grained local features and global contextual information. The proposed approach is evaluated on two benchmark datasets: the Monkeypox Skin Lesion Dataset (MSLD) and the MonkeypoxSkin Image Dataset (MSID). Experimental results demonstrate that the proposed model achieves 99.60% accuracy on the MSLD dataset and 99.19% accuracy on the MSID dataset. The model shows superior performance in terms of accuracy, precision, and robustness, while maintaining computational efficiency suitable for deployment on edge devices. The proposed architecture delivers state-of-the-art performance for monkeypox detection and provides a scalable, efficient, and interpretable solution. This work contributes toward bridging the gap between AI-driven diagnostic tools and real-world clinical applications, enabling early and reliable detection of monkeypox.