Scholarlink

<p>The Monkeypox virus (MPXV), a re-emerging zoonotic threat, continues to pose significant public health challenges due to its symptomatic similarities with diseases such as smallpox and varicella. Timely and accurate diagnosis is critical to containing outbreaks, particularly in regions with limited healthcare infrastructure. To improve robustness and generalization, the approach employs advanced data augmentation methods, including MixUp, CutMix, and GAN-based synthetic image generation, to enhance dataset diversity. Dual-branch feature extraction architecture integrates EfficientNetV2 for capturing fine-grained local details and a Vision Transformer (ViT) for modeling long-range dependencies and global contextual information. The fused feature representation is classified using a custom lightweight Convolutional Neural Network (CNN), delivering binary predictions Monkeypox or Non-Monkeypox with minimal computational cost, enabling deployment on mobile and resource-constrained platforms. Explainable AI (XAI) techniques, specifically Grad-CAM and SHAP, are incorporated to provide visual interpretability and support clinical trust. The resulting deployment-ready framework enables seamless integration into real-world healthcare workflows, facilitating rapid, reliable, and accessible Monkeypox diagnosis. Achieving a detection accuracy of 96.74%, the proposed system outperforms traditional CNN-based models and demonstrates strong potential for real-time deployment using TensorFlow Lite on mobile and edge devices.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hybrid EfficientNetV2–ViT feature extraction with lightweight CNN for explainable and deployable Monkeypox detection

  • 期刊论文

  • 发表日期：

    2025年12月8日

查看全文
  • R. Kaviarasan,
  • P. Sathya Sireesha Bai

摘要

The Monkeypox virus (MPXV), a re-emerging zoonotic threat, continues to pose significant public health challenges due to its symptomatic similarities with diseases such as smallpox and varicella. Timely and accurate diagnosis is critical to containing outbreaks, particularly in regions with limited healthcare infrastructure. To improve robustness and generalization, the approach employs advanced data augmentation methods, including MixUp, CutMix, and GAN-based synthetic image generation, to enhance dataset diversity. Dual-branch feature extraction architecture integrates EfficientNetV2 for capturing fine-grained local details and a Vision Transformer (ViT) for modeling long-range dependencies and global contextual information. The fused feature representation is classified using a custom lightweight Convolutional Neural Network (CNN), delivering binary predictions Monkeypox or Non-Monkeypox with minimal computational cost, enabling deployment on mobile and resource-constrained platforms. Explainable AI (XAI) techniques, specifically Grad-CAM and SHAP, are incorporated to provide visual interpretability and support clinical trust. The resulting deployment-ready framework enables seamless integration into real-world healthcare workflows, facilitating rapid, reliable, and accessible Monkeypox diagnosis. Achieving a detection accuracy of 96.74%, the proposed system outperforms traditional CNN-based models and demonstrates strong potential for real-time deployment using TensorFlow Lite on mobile and edge devices.