Scholarlink

<p>Skin diseases are among the most common health conditions, yet there is a shortage of specialists in this field. Although machine learning has shown significant potential to address this issue, concerns about patient data privacy remain, as training such models typically requires centralized data storage. This paper proposes a new framework that utilizes One-Shot Federated Learning (FL) with dataset distillation to solve the privacy and efficient model training problems of skin disease detection. In this paper, we have implemented FedD3, which combines the privacy benefits of federated learning with the communication efficiency of one-shot communication. This also avoids the iterative model updates across distributed devices, which creates a high bandwidth usage problem. Through our experiments on real skin disease data, this framework gives an accuracy of over 90% for six different types of skin diseases without breaching users’ privacy. The proposed model will not only preserve patients’ privacy but also greatly reduce bandwidth and computational overhead, making it deployable in resource-constrained environments. The study plays a part in the attempt to get closer to the future of privacy-preserving machine learning within medical science.</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

Privacy-Preserving Skin Disease Detection Using One-Shot Federated Learning Approach

  • 期刊论文

  • 发表日期：

    2025年12月11日

查看全文
  • Tasnim Ur Rahaman Anas,
  • Qaiser Razi,
  • Sparsh Bajoria,
  • Vikas Hassija,
  • GSS Chalapathi

摘要

Skin diseases are among the most common health conditions, yet there is a shortage of specialists in this field. Although machine learning has shown significant potential to address this issue, concerns about patient data privacy remain, as training such models typically requires centralized data storage. This paper proposes a new framework that utilizes One-Shot Federated Learning (FL) with dataset distillation to solve the privacy and efficient model training problems of skin disease detection. In this paper, we have implemented FedD3, which combines the privacy benefits of federated learning with the communication efficiency of one-shot communication. This also avoids the iterative model updates across distributed devices, which creates a high bandwidth usage problem. Through our experiments on real skin disease data, this framework gives an accuracy of over 90% for six different types of skin diseases without breaching users’ privacy. The proposed model will not only preserve patients’ privacy but also greatly reduce bandwidth and computational overhead, making it deployable in resource-constrained environments. The study plays a part in the attempt to get closer to the future of privacy-preserving machine learning within medical science.