Lightweight LLM-Based Anomaly Detection Framework for Securing IoTMD Enabled Diabetes Management Control Systems

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Medical Devices

Background:

• Implantable Internet of Things Medical Devices (IoTMD) enhance chronic disease management through continuous monitoring.
• IoTMD systems face significant security risks from cyber threats, potentially compromising patient data and device function.
• Securing IoTMD is crucial for reliable healthcare delivery.

Purpose of the Study:

• To evaluate lightweight Large Language Models (LLMs) for anomaly detection in IoTMD-enabled diabetes management control systems (DMCS).
• To assess the performance of fine-tuned LLMs against traditional deep learning models for cybersecurity in medical devices.

Main Methods:

• Fine-tuning lightweight LLMs (LLaMA 3.2 1B-Instruct, GPT-2, Phi-1, Gemma 2B-Instruct) using Low-Rank Adaptation (LoRA).
• Anomaly detection in DMCS using transformer-based LLMs.
• Comparative analysis against traditional models (IL-MLP, IL-CNN, FL-MLP, FL-CNN).

Main Results:

• LLaMA 3.2 1B-Instruct, fine-tuned with LoRA, achieved 99.91% accuracy, 100% precision, and 0% false positive rate.
• Transformer-based LLMs demonstrated superior adaptability and robustness in anomaly detection compared to other models.
• The study confirmed the effectiveness of LLMs in identifying threats to IoTMD systems.

Conclusions:

• Fine-tuned LLMs offer a robust solution for enhancing the cybersecurity of IoTMD systems.
• LLM-based anomaly detection strengthens the reliability and safety of implantable medical devices.
• This approach paves the way for more secure and dependable medical technologies in healthcare.