Scholarlink

<p>Magnetic Resonance Imaging (MRI) is a crucial tool in medical diagnostics, yet reconstructing high-quality images from under-sampled k-space data poses significant challenges. This study introduces Federated Adversarial MRI Enhancement (FAME), a novel framework combining Federated Learning (FL) with Generative Adversarial Networks (GANs) to enhance MRI reconstruction while maintaining patient privacy. FAME utilizes a hybrid model aggregation strategy that dynamically weights updates from local generators, ensuring a balanced contribution based on dataset size and quality. Each local generator is trained on-site-specific data, while a global discriminator evaluates and refines the aggregated updates to improve image quality. FAME addresses key issues in medical imaging, including data privacy, model generalization, and robustness, by integrating advanced GAN architectures such as multi-scale convolutions, attention mechanisms, and Graph Neural Networks (GNNs). Differential privacy and secure aggregation protocols are implemented to protect sensitive data during training. Extensive experiments using the fastMRI Brain and Knee datasets, along with the BraTS 2020 and IXI dataset, show that FAME outperforms existing models, achieving superior PSNR and SSIM values. This decentralized framework offers scalable, privacy-preserving MRI reconstruction, making it a promising solution for diverse clinical applications.</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

FAME: A Federated Adversarial Learning Framework for Privacy-Preserving MRI Reconstruction

  • 期刊论文

  • 发表日期：

    2025年1月25日

查看全文
  • Shahzad Ahmed,
  • Jinchao Feng,
  • Javed Ferzund,
  • Muhammad Yaqub,
  • Muhammad Usman Ali,
  • Malik Abdul Manan,
  • Abdul Raheem

摘要

Magnetic Resonance Imaging (MRI) is a crucial tool in medical diagnostics, yet reconstructing high-quality images from under-sampled k-space data poses significant challenges. This study introduces Federated Adversarial MRI Enhancement (FAME), a novel framework combining Federated Learning (FL) with Generative Adversarial Networks (GANs) to enhance MRI reconstruction while maintaining patient privacy. FAME utilizes a hybrid model aggregation strategy that dynamically weights updates from local generators, ensuring a balanced contribution based on dataset size and quality. Each local generator is trained on-site-specific data, while a global discriminator evaluates and refines the aggregated updates to improve image quality. FAME addresses key issues in medical imaging, including data privacy, model generalization, and robustness, by integrating advanced GAN architectures such as multi-scale convolutions, attention mechanisms, and Graph Neural Networks (GNNs). Differential privacy and secure aggregation protocols are implemented to protect sensitive data during training. Extensive experiments using the fastMRI Brain and Knee datasets, along with the BraTS 2020 and IXI dataset, show that FAME outperforms existing models, achieving superior PSNR and SSIM values. This decentralized framework offers scalable, privacy-preserving MRI reconstruction, making it a promising solution for diverse clinical applications.