PhysioKey: Edge-AI-Driven Physiological Key Agreement for Secure Body Area Networks

Area of Science:

• Biomedical Engineering
• Cybersecurity
• Machine Learning

Background:

• Body area networks (BANs) face security challenges due to resource-constrained sensor nodes.
• Conventional cryptography and pre-shared keys are unsuitable for BANs and clinical workflows.
• Need for secure, plug-and-play intra-body communication solutions.

Purpose of the Study:

• Introduce PhysioKey, a TinyML-based framework for secure key agreement in BANs.
• Derive symmetric session keys from physiological signals without pre-shared secrets.
• Enable secure intra-body communication for resource-limited medical devices.

Main Methods:

• Utilized a lightweight 1D-CNN (6320 parameters, INT8-quantized) for feature extraction from ECG and PPG signals.
• Employed fuzzy commitment with BCH error-correcting codes for key reconciliation.
• Implemented patient-level 5-fold cross-validation on PTB-XL and BIDMC datasets.

Main Results:

• Achieved an Equal Error Rate (EER) of 7.8%±0.8% on dual-ECG data (PTB-XL).
• Reduced cross-modal EER to 30.6%±1.2% using a dual-encoder architecture (BIDMC: ECG + PPG).
• Standalone PhysioKey provides 7-24 effective bits; hybrid mode with ECDH offers 128-bit security.

Conclusions:

• PhysioKey provides a viable TinyML solution for secure key agreement in BANs.
• Hybrid PhysioKey + ECDH offers robust security with physical on-body authentication.
• Standalone PhysioKey is suitable for energy-constrained scenarios, offering a 27x advantage over ECDH.