A Hybrid Mutual Authentication Approach for Artificial Intelligence of Medical Things

Mian Ahmad Jan, Wenjing Zhang, Aamir Akbar, Houbing Song, Rahim Khan, Samia Allaoua ChellougIEEE Internet of Things Journal 2024 · DOI

Introduction

The Artificial Intelligence of Medical Things (AIoMT) represents the convergence of smart wearable sensors and advanced data analytics. While these systems enable real-time health monitoring, they are inherently vulnerable. Resource-constrained devices like heart rate monitors and glucose sensors lack the processing power for traditional, heavy encryption, making them prime targets for eavesdropping and Man-in-the-Middle (MITM) attacks. This paper proposes a lightweight, hybrid mutual authentication scheme to bridge the gap between robust security and high-performance AI analytics.

Figure 1: Overview of the AIoMT ecosystem, including wearable devices, gateway modules, and the AI-enabled cloud.

Proposed Methodology

The authors introduce a three-phase approach designed for efficiency and device mobility in clinical settings.

1. Offline Registration

Before deployment, devices and servers exchange MAC addresses in a secure environment. Devices are categorized as "member-devices" (under a specific server's care) or "non-member-devices" (legitimate devices registered to other nearby servers).

2. Authentication Phase

When a device initiates data transmission, it undergoes a lightweight handshake. This phase utilizes AES-128 bit encryption for the payload and MAC-address-based verification for identity confirmation. A unique localization function also calculates the physical distance between the device and the gateway to filter out remote intruders.

Figure 2: The mutual authentication handshake between a mobile device (Ci) and a server (Sj).

3. Communication and Mobility

Once authenticated, the system supports device mobility (e.g., a patient moving between hospital wards) through a "handover" mechanism. This allows a new server to verify a device's legitimacy without requiring a power-intensive full re-registration.

AI-Driven Anomaly Detection

Beyond security, the framework integrates authenticated data into Deep Autoencoders and Recurrent Neural Networks (RNNs). These models establish a baseline for "normal" physiological patterns and flag deviations as potential health anomalies, ensuring that the data used for clinical decisions is both secure and intelligent.

Performance Evaluation

The scheme was evaluated using the OMNET++ simulation platform, modeling up to 1,000 devices. Key results include:

  • Efficiency: The protocol requires only four messages and minimal computational overhead (4 hash and random nonce operations), significantly lower than competing 3-factor authentication schemes.
  • Communication Overhead: The process uses only 4,096 bits, compared to over 30,000 bits in some existing models.
  • Security Robustness: The approach successfully defended against six major attack vectors, including MITM and eavesdropping, while maintaining high throughput in dense environments.

Conclusion

By leveraging hardware-native MAC addresses alongside AES encryption, this hybrid approach provides mutual authentication with minimal resource consumption. Future work will focus on real-world hospital deployments and the potential integration of Large Language Models (LLMs) to enhance the interpretability of AI-generated health alerts.

Summary updated

BibTeX
@article{jan2024hybrid,
  author = {Mian Ahmad Jan and Wenjing Zhang and Aamir Akbar and Houbing Song and Rahim Khan and Samia Allaoua Chelloug},
  title = {{A Hybrid Mutual Authentication Approach for Artificial Intelligence of Medical Things}},
  journal = {IEEE Internet of Things Journal},
  year = {2024},
  doi = {10.1109/jiot.2023.3317292},
}
Figure