Federated Autoencoder Model for Secure Medical Image Analysis With Privacy Preservation and Assurance.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Edge Computing
• Data Privacy

Background:

• Medical imaging analysis on edge devices faces challenges in balancing computational demands with patient privacy and data security.
• Existing methods often struggle to maintain high accuracy while implementing robust privacy-preserving mechanisms.

Purpose of the Study:

• To present U-NeTrans, a novel federated autoencoder model designed for secure and private medical image reconstruction on edge devices.
• To enhance the capabilities of edge devices for medical imaging analysis without compromising patient confidentiality.

Main Methods:

• Developed U-NeTrans, a federated autoencoder model incorporating random masking for training complexity and partial data utilization.
• Implemented a privacy-preserving mechanism where the encoder processes visible patches and the decoder uses encoded data for image reassembly.
• Integrated auxiliary reconstruction tasks and contrastive loss to improve the representation of high-order features in medical images.

Main Results:

• U-NeTrans demonstrated superior performance compared to state-of-the-art methods in edge-based medical image analysis.
• Achieved high accuracy (98.97%), precision (98.68%), sensitivity (98.73%), and AUROC (99.19%) in experimental evaluations.
• Effectively maintained patient security and privacy throughout the image analysis process.

Conclusions:

• U-NeTrans offers a significant advancement in secure and private medical image analysis on edge devices.
• The model has broad applications, particularly in chest X-ray analysis, with the potential to enhance edge healthcare capabilities.
• The proposed method effectively balances precise medical image analysis with stringent patient privacy requirements.