GAN-based patient information hiding for an ECG authentication system.

Area of Science:

• Biometrics
• Machine Learning
• Cardiology

Background:

• Biometric authentication, using unique physiological patterns like fingerprints, is common for personal identification.
• Physiological data, such as electrocardiogram (ECG) signals, can contain sensitive health information, including arrhythmia patterns.
• Current authentication methods on mobile devices require robust security to protect personal health information.

Purpose of the Study:

• To develop a lightweight multi-task deep learning model for simultaneous personal identification and arrhythmia detection using ECG signals.
• To ensure the privacy of sensitive health data within biometric authentication systems.
• To create an efficient model suitable for implementation on resource-constrained mobile devices.

Main Methods:

• A multi-task neural network was designed to process ECG signals for both individual identification and arrhythmia pattern recognition.
• The model was trained and evaluated for its performance in authentication and health monitoring tasks.
• Computational efficiency and model size were compared against single-task models based on parameter count.

Main Results:

• The multi-task model achieved comparable performance to single-task models while utilizing approximately 20,000 fewer parameters.
• Demonstrated the feasibility of integrating personal identification and health monitoring within a single, efficient deep learning architecture.
• The study confirmed the model's suitability for mobile device deployment due to its reduced size and computational demands.

Conclusions:

• A novel multi-task deep learning approach offers an efficient solution for secure personal authentication using ECG data.
• This method effectively protects sensitive health information by integrating it within a secure identification framework.
• The developed lightweight model presents a viable option for enhancing security and functionality on mobile devices.