A Heterogeneous Neural Network With Multi-Modal Butterfly Attractors for Secure Medical Image Transmission

This work proposes a novel minimal hybrid memristive neural network (HT-MNN) for secure grayscale medical image transmission in the Internet of Medical Things (IoMT). The HT-MNN integrates a Hopfield neuron and a Tabu learning neuron via dual memristive couplings, generating rich multi-modal chaotic butterfly attractors including multi-wing and initial-boosted coexisting structures. The attractors’ number and spatial distribution are flexibly tunable via memristive parameters and initial conditions for controllable dynamical complexity, with circuit implementation validated via PSIM simulations. Leveraging its high-entropy chaotic sequences, we design a medical image encryption scheme combining row–column permutation and multi-stage diffusion. For 512×512 clinical grayscale images, performance evaluation shows encrypted images achieve Peak Signal-to-Noise Ratio (PSNR) below 8.5 dB, Mean Squared Error (MSE) over 9200, information entropy of 7.9993 (near the ideal 8.0) and an ultra-large 2816key space. Compared with state-of-the-art medical image encryption methods, the scheme achieves lower encrypted-image PSNR and higher throughput of 11.27 Mbps. With its balanced lightweight design, high-grade security, and strong antinoise robustness, the HT-MNN is well-suited for fast and secure IoMT clinical scenarios such as bedside portable imaging.