A joint data-driven compressive sensing and latent domain hyperchaotic encryption framework for medical images

Jia Y., Guo J., ERKAN U., TOKTAŞ A.

EUROPEAN PHYSICAL JOURNAL PLUS, cilt.141, sa.9, 2026 (SCI-Expanded)

Özet

To address the challenges of low compression ratio reconstruction quality degradation, potential data leakage, and ciphertext error propagation in medical image transmission under bandwidth constraints and noise interference, existing methods typically design compressed sensing and chaotic encryption separately, making it difficult to simultaneously achieve low compression ratio reconstruction, latent domain data protection, and transmission robustness. Therefore, this paper proposes a deep learning compression-chaotic encryption framework, SE-AMCSNet-LCF. The compression-excited adaptive medical compressed sensing network (SE-AMCSNet) generates compact latent measurements through trainable convolutional sampling and completes image reconstruction using sub-pixel mapping and channel attention residual structures. The Latent Domain Chaos Framework (LCF) combines plaintext-related hyperchaotic initialization, independent bit-plane scrambling, and nonlinear two-dimensional FIR diffusion to encrypt quantized latent measurements. Experimental results show that even with a compression ratio of 0.04, the proposed framework can achieve an average peak signal-to-noise ratio (PSNR) of 32.235 dB and a sequence similarity index (SSIM) of 0.8466, while also exhibiting good statistical security. These results demonstrate that the framework provides an effective solution for the efficient and secure transmission of medical images in bandwidth-constrained environments.