Deep learning-based encryption scheme for medical images using DCGAN and virtual planet domain.

Area of Science:

• Medical Imaging
• Cybersecurity
• Cryptography

Background:

• Medical image security is critical for patient confidentiality and data integrity.
• Existing encryption methods may not sufficiently address sophisticated threats like tampering and adversarial attacks.

Purpose of the Study:

• To present a novel encryption technique for medical images.
• To enhance the security and confidentiality of sensitive medical data.

Main Methods:

• Integration of Deep Convolutional Generative Adversarial Networks (DCGAN) and Virtual Planet Domain (VPD) for image encryption.
• Generation of encryption keys using a Deep Learning (DL) framework, timestamp, nonce, and 1-D Exponential Chebyshev map (1-DEC).

Main Results:

• Experimental results demonstrate the technique's efficacy against unauthorized access, tampering, and adversarial attacks.
• National Institute of Standards and Technology (NIST) SP 800-22 tests confirm the randomness of keys and encrypted images.
• The algorithm shows robustness against key sensitivity, noise, cropping, and adversarial attacks, with efficient time and key space complexity.

Conclusions:

• The proposed encryption algorithm offers high security and reliability for medical images, as evidenced by Information Entropy, correlation coefficient, MSE, PSNR, NPCR, and UACI metrics.
• Statistical analyses and comparisons with existing algorithms indicate the proposed method is suitable for practical implementation in medical settings.