BF2GAN: A Fused Federated Learning-Based Biometric Enabled Deep Learning Approach for Secure Medical Image Sharing in...

Area of Science:

• Medical Imaging
• Cybersecurity
• Artificial Intelligence

Background:

• Medical images contain sensitive patient data requiring robust security measures.
• Existing encryption methods face challenges with complexity, compression, and real-time performance.
• Unauthorized access to medical images poses significant privacy risks.

Purpose of the Study:

• To propose a novel method, Biometric Fused Federated Generative Adversarial Network (BF2GAN), for enhanced medical image security.
• To address limitations of current encryption schemes in medical image sharing.
• To improve data integrity and reduce data exposure risks during collaborative training.

Main Methods:

• Implementation of federated learning (FL) for decentralized, collaborative model training.
• Integration of biometric information for user verification and secure key generation.
• Utilizing Generative Adversarial Networks (GANs) within a federated framework.

Main Results:

• BF2GAN achieved efficient encryption (3.08s) and decryption (3.9s) times.
• High image quality was maintained, evidenced by a Structural Similarity Index Measure (SSIM) of 0.975 and Feature Similarity Index (FSI) of 0.82.
• The method demonstrated strong performance with a Peak Signal-to-Noise Ratio (PSNR) of 58.38 dB and minimal memory usage (280.87 KB).

Conclusions:

• BF2GAN offers a simplified, secure encryption process for medical images, reducing reliance on complex key management.
• The method enhances data integrity and privacy through decentralized training and biometric authentication.
• BF2GAN presents a superior alternative to state-of-the-art methods for secure medical image sharing.