Scholarlink

This research presents a privacy-preserving framework for brain tumor classification of MRI images using Federated Learning (FL), Transfer Learning (TL) on a DenseNet model, and Somewhat Homomorphic Encryption (SHE). While deep learning models have shown promise in medical image analysis, privacy concerns over sensitive data limit their use. To address this, we combine SHE for secure data processing with FL, allowing decentralized model training while keeping MRI data local. The DenseNet model, enhanced through Transfer Learning, leverages pre-trained weights to improve classification accuracy with limited data. Our system ensures that data privacy is preserved throughout the training process while achieving high performance in brain tumor detection. The results show that the proposed framework effectively classifies MRI images while maintaining confidentiality with accuracy of 94.94% offering a secure and scalable solution for medical image classification in a distributed environment. This approach holds potential for widespread use in privacy-sensitive medical applications.

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

Leveraging FedSHE-Based FedAVG for Efficient Privacy-Preserving Federated Learning and Secure Model Aggregation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Rashmi Ranjan Maharana,
  • Rashmi Ranjan Sahoo,
  • Dinesh Kumar Dash

摘要

This research presents a privacy-preserving framework for brain tumor classification of MRI images using Federated Learning (FL), Transfer Learning (TL) on a DenseNet model, and Somewhat Homomorphic Encryption (SHE). While deep learning models have shown promise in medical image analysis, privacy concerns over sensitive data limit their use. To address this, we combine SHE for secure data processing with FL, allowing decentralized model training while keeping MRI data local. The DenseNet model, enhanced through Transfer Learning, leverages pre-trained weights to improve classification accuracy with limited data. Our system ensures that data privacy is preserved throughout the training process while achieving high performance in brain tumor detection. The results show that the proposed framework effectively classifies MRI images while maintaining confidentiality with accuracy of 94.94% offering a secure and scalable solution for medical image classification in a distributed environment. This approach holds potential for widespread use in privacy-sensitive medical applications.