AFM-DViT: A framework for IoT-driven medical image analysis

Abstract

With the rise of Internet of Things (IoT) in healthcare, automated medical image analysis has become essential for real-time disease detection. However, current models face limitations in handling diverse datasets and ensuring privacy across distributed systems. To address these challenges, we propose the AFM-DViT model, which integrates adaptive federated learning with Vision Transformer, significantly enhancing diagnostic accuracy and efficiency in IoT-based settings. Our framework not only improves detection capabilities but also effectively addresses critical issues related to data privacy and heterogeneity in medical imaging. Experimental results demonstrate that AFM-DViT outperforms state-of-the-art methods by achieving an AUROC of 0.841 and sensitivity of 0.925 on the ChestX-ray14 dataset, alongside a sensitivity of 0.888 with an AUC of 0.905 on the LUNA16 dataset. These results highlight the model’s robust detection accuracy while maintaining data privacy. The AFM-DViT model offers an effective solution for secure and efficient medical image analysis in IoT-enabled environments.

Keywords

• Adaptive federated learning
• Vision transformer
• IoT
• Medical image analysis
• Privacy preservation