Phase-domain Deep Patient-ECG Image Learning for Zero-effort Smart Health Security

Area of Science:

• Biometrics
• Cybersecurity
• Medical Informatics

Background:

• Smart health relies on ubiquitous sensors, generating vast data vulnerable to security breaches.
• Wearable devices have power constraints limiting complex security protocols.
• Existing methods for ECG-based identification face challenges with diverse morphologies and manual feature engineering.

Purpose of the Study:

• To develop a zero-effort framework for user identification using electrocardiogram (ECG) signals.
• To address challenges in ECG biometric identification, including signal diversity and feature extraction.
• To enhance smart health security without requiring additional hardware.

Main Methods:

• A phase-domain deep patient-ECG image learning framework was investigated.
• Phase-domain transformation enabled blind signal segmentation without heartbeat localization.
• ECG trajectories were converted into images for deep convolutional neural network analysis.

Main Results:

• The framework achieved an accuracy of 97.2% on two patient-ECG databases.
• Outperformed state-of-the-art methods in generalization ability and performance.
• Demonstrated effective user identification using ECG biometrics.

Conclusions:

• The proposed zero-effort framework offers a robust solution for patient-ECG biometric user identification.
• Generalizable blind signal segmentation and deep feature learning are key to enhancing smart health security.
• This approach is crucial for the security of the Internet of Medical Things and big medical data.