Plausible Deniable Medical Image Encryption by Large Language Models and Reversible Content-Aware Strategy.

Area of Science:

• Computer Science
• Medical Imaging
• Cryptography

Background:

• Healthcare systems face significant security risks, with data loss potentially causing severe harm to patients and institutions.
• DNA encoding presents a promising solution for medical image encryption due to its speed, parallelism, minimal storage, and robust security.

Purpose of the Study:

• To propose an advanced medical image encryption technique by integrating Large Language Models (LLMs) with DNA encoding.
• To enhance the security and efficiency of medical image data protection against privacy breaches.

Main Methods:

• The proposed method features an LLM enhancing module to compress low-entropy medical image pixels into compact, probabilistic signals, adding a layer of defense.
• A content-aware permutation and diffusion module further secures data by breaking pixel correlations and increasing cryptographic complexity via DNA-based diffusion.

Main Results:

• Experiments on ChestXray-14, COVID-CT, and fcon-1000 datasets demonstrated superior performance compared to existing methods.
• The LLM-enhanced DNA encoding method showed significant improvements in sensitivity, correlation, and entropy metrics.

Conclusions:

• The LLM-enhanced DNA encoding method provides a highly effective and secure approach for medical image encryption.
• This technique offers a robust solution for safeguarding sensitive medical data in healthcare systems.