Multi-scroll chaotic attractors in a memristive-cyclic Hopfield neural network for ECG signal encryption in telemedicine.

The expansion of telemedicine requires robust protection of physiological data. To meet this need, this paper introduces an electrocardiogram (ECG) encryption framework based on a six-dimensional memristive-cyclic Hopfield neural network (MC-HNN) combined with an adaptive non-uniform partition scheme. The MC-HNN produces chaotic sequences with controllable multi-scroll attractors, while an adaptive cubic-spline partition method captures the nonstationary characteristics of ECG signals to derive signal-dependent initial conditions. These sequences drive a permutation and double-diffusion encryption process. Experiments on the MIT-BIH, PTB-XL, CinC Challenge 2017, and CPSC2018 datasets yield a Number of Sample Change Rate (NSCR) of 100%, an average Unified Average Change Intensity (UACI) of 33.42%, near-zero correlation, entropy approaching 8 bits, and uniform histograms. The scheme resists differential, statistical, and chosen-ciphertext attacks. Encryption of 10-second ECG segments requires less than 0.05 s, demonstrating computational efficiency suitable for practical telemedicine deployment.