A semantic bit-plane based three-layer encryption framework for secure medical images

Secure transmission and storage of medical images is critical as patient information is sensitive and the trend toward digital healthcare, telemedicine and cloud based medical systems is growing rapidly. To address the increased challenges of security and privacy, a novel three layer medical image encryption algorithm is proposed providing a high level of security as well as providing lossless reconstruction for accurate diagnosis when using medical images. This algorithm decomposes the medical image according to bit-planes, into three semantic layers that contain structural, detail, and fine information, making it possible for medical images to be differentiated by level of protection based on the sensitivity of the information contained therein. A separate encryption process is applied for each of the three semantic layers, each using a different cryptographic key created from a single master key generated from an SHA-256 based key derivation process. This will eliminate key exchange overhead and improve the sensitivity of the keys being used. The key dependent substitution, block level diffusion, keystream encryption, and pixel chaining operations used for encrypting the layers provide confusion and diffusion within the system while maintaining the efficiency of computation. The three semantic layers that were previously encrypted are combined and processed with a global plaintext dependent diffusion operation, providing resistance to statistical and differential attacks. The results were evaluated experimentally across several imaging modalities demonstrating strong performance achieving an average entropy of 7.9998 indicating high randomness and uniform pixel distribution, near zero correlation confirming effective removal of spatial relationships between adjacent pixels, an average Number of Pixel Change Rate (NPCR) of 99.61 percent showing the percentage of pixels that change significantly when a single pixel in the original image is modified, and Unified Average Changing Intensity (UACI) of 33.47 percent indicating the average intensity difference between the original and encrypted images, and infinite Peak Signal to Noise Ratio (PSNR) confirming perfect lossless reconstruction.