Deep-KEDI: Deep learning-based zigzag generative adversarial network for encryption and decryption of medical images

Area of Science:

• Medical Imaging Security
• Deep Learning Applications
• Cryptography

Background:

• Medical imaging generates sensitive patient data, necessitating robust security measures for privacy and legal compliance.
• Internet transmission of clinical images poses risks to patient confidentiality and hospital liability.
• Advancements in medical imaging require enhanced data security protocols.

Purpose of the Study:

• To develop a novel deep learning-based network, Deep-KEDI, for secure key generation in medical image encryption.
• To ensure the confidentiality and integrity of sensitive patient data within medical images.
• To provide a secure method for encrypting and decrypting medical images.

Main Methods:

• Medical images are pre-processed with speckle noise using discrete ripplet transform for added security.
• A zigzag generative adversarial network (ZZ-GAN) is employed within the Deep-KEDI model to generate secret keys.
• The ZZ-GAN generates keys for encryption/decryption using XOR operations and three zigzag patterns (vertical, horizontal, diagonal).

Main Results:

• The ZZ-GAN successfully encrypts and decrypts images using generated secret keys and XOR operations.
• Speckle noise is effectively removed post-decryption to reconstruct the original medical image.
• The Deep-KEDI model demonstrated effective key generation for secure medical image processing.

Conclusions:

• The Deep-KEDI model provides a secure and effective method for medical image encryption.
• The generated secret keys exhibit high information entropy (7.45), suitable for securing sensitive medical data.
• This deep learning approach enhances the security and privacy of medical images during transmission and storage.