A deep learning-orchestrated garlic routing architecture for secure telesurgery operations in healthcare 4.0

Recently, the Internet of Things (IoT) has attracted different real-time services, predominantly in the healthcare domain. One such real-time IoT-based application is telesurgery, where surgeons remotely transmit surgery instructions to a robotic arm, enabling it to conduct surgical procedures on patients. Since these surgical instructions use conventional wireless networks, they can leveraged by the attackers to manipulate them and manoeuvre the entire telesurgery application. Therefore, in this paper, we used emerging technologies, such as Artificial Intelligence (AI), garlic routing (GR) networks, and blockchain, to propose an AI- and GR-based secure data instruction architecture for telesurgery applications in the healthcare 4.0 domain. A standard sensor dataset is utilized to train different AI algorithms, such as Long Short Term Memory (LSTM) and Gated Recurrent Neural Networks (GRU), for classifying malicious and non-malicious telesurgery data. Further, the non-malicious data is forwarded to the GR network that provides an end-to-end encrypted tunnel using ElGamal and Advanced Encryption Standard (AES). ElGamal encryption encrypts the session tags for each telesurgery data relayed between surgeons and the robotic arm. The tags are stored in the immutable blockchain nodes to avoid data tampering attacks that strengthen the legitimacy of the garlic routers. Among both, the GRU outperforms with test accuracy 96.89%, precision 97.32%, recall 96.46%, F1 score 96.86%, and training loss 3%. Furthermore, the telesurgery data is transmitted via an AES-based outbound tunnel and received via an AES-based inbound tunnel, offering robust security against the security threats associated with the telesurgery application. To improve the network performance, we used essential characteristics (ultra-low latency, high speed, and high reliability) of the 5G network interface between each layer of the proposed architecture. The proposed architecture is evaluated using different evaluation metrics, such as statistical analysis (training accuracy, training loss, optimizer performance, activation function performance), data compromisation rate (0.346), network throughput (1.44 Mbps), error rate, and latency comparison.