Enhancing Patient Privacy in IoT-Enabled Intelligent Systems: A Deep and Broad Learning-Based Efficient Encryption Network
Xiaoqiang Zhu, Jiqiang Liu, Dalin Zhang, Lingkun Li, Nan Wang · 2024
The rapid development of the Internet of Things (IoT) is enabling a wide range of applications in intelligent medical systems. Among others, medical imaging equipment produces sensitive user privacy information, however, current solutions from academia and industry often neglect the importance of secure communication mechanisms. There are open research challenges such as low real-time processing and poor security, even when using cryptography. This paper proposes EDBNet, an efficient encryption network based on deep and broad learning to improve patient privacy for medical images. To be specific, a four-layer convolutional neural network is employed to extract the horizontal and vertical factors and utilize broad learning to guide the training model to obtain two feature matrices. The training process includes pre-training and fine-tuning, with the open-source COVID-CT-Dataset enabling dual-stream encryption. To further enhance ciphertext image security in a privacy-protected environment, chaotic cryptography is utilized to consummate the encryption network, which includes scrambling and diffusion combing with the SHA3-256 algorithm. The proposed EDBNet is evaluated by extensive experiments, which show that it outperforms several state-of-the-art algorithms, such as the average cipher entropy of 7.9971, encryption quality of 248, and encrypted/decrypted time of around 1 second.