Dynamic memdiode–driven Hopfield neural network for 3D CT encryption in IoMT

Chaos-based neural networks with memristive behavior are increasingly used for secure communications and medical data protection in Internet-of-Medical-Things (IoMT) systems. In this study, inspired by compact dynamic memdiode models, we propose a dynamic memdiode model–driven heterogeneous memristive Hopfield neural network (DMM–Het–MHNN). The proposed system connects three continuous-time neurons through a single dynamic memdiode, forming a four-dimensional chaotic core. Its dynamics are analyzed to identify chaotic regimes suitable for keystream generation. Using this chaotic core, a two-round 3D volume permutation–diffusion encryption scheme is developed for Hounsfield Units (HU)-calibrated 16-bit computed tomography (CT) volumes obtained from Digital Imaging and Communications in Medicine (DICOM) data. All encryption operations are performed in the integer domain, allowing exact recovery with the correct secret key. Tests on four public clinical CT datasets show near-ideal 16-bit entropy, low neighboring-voxel correlation, and nearly uniform ciphertext histograms. In addition, the chaotic core is implemented on an STM32F303ZE microcontroller, where measured phase portraits closely match simulation results, confirming embedded feasibility.