MonkeyPix: A Vision Transformer Solution for Efficient Monkeypox Detection in Low-resource Settings

Monkeypox is a rapidly spreading infectious skin disease that presents significant challenges for diagnosis, especially in low-resource environments. While traditional diagnostic methods such as clinical observation and PCR testing are highly accurate, they often require expensive equipment and skilled personnel, making them less accessible in resource-limited regions. To address these issues, this study introduces MonkeyPix, a lightweight Vision Transformer (ViT)-based model optimized for rapid and accurate monkeypox diagnosis. MonkeyPix incorporates innovative features, including pixel-wise self-attention and the Shifted Window mechanism derived from the Swin Transformer, to reduce the model size by approximately 80% without compromising diagnostic performance. This model achieves high efficiency, enabling effective diagnosis even in regions with minimal healthcare infrastructure. Experimental results demonstrate that MonkeyPix not only outperforms conventional convolutional neural network (CNN)-based models but also significantly reduces computational and memory requirements. Moreover, MonkeyPix shows potential for broader applications, including the diagnosis of various other skin diseases. By leveraging its lightweight architecture, the model facilitates the integration of advanced AI tools into under-resourced medical systems, providing a scalable and cost-effective solution to global health challenges. Future research will focus on validating the model's performance in real-world clinical settings and expanding its use to a wider range of diseases, ultimately contributing to improved public health outcomes.