Privacy-Preserving for Medical Images Using Cryptosystem and Convolutional Autoencoder
Charles Norgbe, Mahdi Madani, El‐Bay Bourennane · 2025
The adoption of cloud services for storing and sharing personal information, especially sensitive medical images, continues to rise, and concerns regarding data security and user privacy have become more significant. This paper addresses this challenge by proposing a novel compressed-domain encryption framework that simultaneously achieves image compression and cryptographic protection of user data. The primary objective is to enable the secure transmission of medical images while reducing data size and preserving visual quality. The proposed system integrates an autoencoder-based compression model with Advanced Encryption Standard - Galois/Counter Mode (AES-GCM) encryption to ensure data confidentiality in untrusted environments. A composite loss function is introduced to optimise performance, improving perceptual similarity between original and reconstructed images. Experimental results demonstrate high reconstruction quality, achieving a Structural Similarity Index Measure (SSIM) of 98.7 % and successful decryption and efficient compression. The system also demonstrates effective decryption and compression performance, validating its suitability for secure medical image transmission and storage in cloud-based healthcare applications.