MAN-C: A masked autoencoder neural cryptography based encryption scheme for CT scan images.

Area of Science:

• Medical Imaging
• Cryptography
• Computer Vision

Background:

• Secure sharing of medical images is crucial for patient data confidentiality.
• Existing methods may face challenges in maintaining image integrity during transmission.

Purpose of the Study:

• To propose MAN-C, a Masked Autoencoder Neural Cryptography based encryption scheme for secure medical image sharing.
• To leverage masked autoencoders and neural cryptography for enhanced security and image reconstruction.

Main Methods:

• Utilizing masked autoencoders, originally for self-supervised learning in computer vision, as an encryption-decryption mechanism.
• Integrating autoencoders with neural cryptography, including a tree parity machine and Shamir Scheme, for secret image sharing.
• Employing CT scans from The Cancer Imaging Archive (TCIA) for evaluation.

Main Results:

• The MAN-C scheme demonstrates improved Root Mean Square Error (RMSE) values compared to existing techniques.
• Comparable correlation values between input and output images were achieved.
• The method effectively recovers image loss caused by noise during secret sharing.

Conclusions:

• MAN-C offers a secure and efficient method for sharing medical images.
• The combination of masked autoencoders and neural cryptography provides a robust encryption solution.
• This approach enhances confidentiality while maintaining image quality and integrity.