Multi-scale dilated depthwise separable convolutional network driven secure medical image transmission in wireless multimedia sensor networks

Secure and reliable biomedical image transfer is a serious problem in healthcare-focused wireless multimedia sensor networks (WMSN), especially in dealing with sensitive modalities like ultrasound imaging. To solve this, this manuscript introduces a strong system for secure ultrasound image transmission that employs electrocardiogram (ECG) signals as cover media. The procedure starts with key generation, in which a light-weight cryptosystem integrating the Henon chaotic map (HcM), Brownian motion (BM), and Chen's chaotic system (CcS) generates highly random and unpredictable keys. These keys are then utilized to drive a Polymorphic Advanced Encryption Standard (Poly-AES) module, providing robust encryption of ultrasound images and ensuring confidentiality protection against cryptanalytic attacks. After encrypting, the images are embedded in the ECG signals by implementing an Activation-Aware Singular Value Decomposition (AA-SVD) approach. The embedding procedure adapts its strength depending on activation functions to achieve an optimal balance between invisibility and resilience. After embedding the encrypted images into the ECG signals, the communication step is accomplished through the use of a Multi-scale Dilated Depthwise Separable Convolutional Neural Network (MS-DDSCN) that can fuse multi-scale contextual information without sacrificing computational cost, therefore resisting any channel distortion. On the other hand, the decryption procedure involves the reverse AA-SVD procedure and chaotic-polymorphic AES key decryption to ensure proper decoding of the original ultrasound images. Finally, the system performance will be validated via extensive experimental testing on the publicly available breast ultrasound image database and ECG signals (MIT-BIH Arrhythmia Dataset). Performance is gauged using the image quality measures. PSNR, SSIM, NCC, robustness parameters such as BER, correlation coefficient, and security parameters like entropy, NPCR, UACI, throughput, and energy are considered to validate the feasibility of the presented model in applications within the field of medicine-based IoT. It can be seen from the experiment that the proposed model achieves PSNR = 42.7 dB and SSIM = 0.98, showing that high-quality image recovery is possible in the proposed system, and the result for NPCR = 99.62% signifies good defense against differential attacks.