Enhancing biometric identification using 12-lead ECG signals and graph convolutional networks.

Area of Science:

• Biometrics
• Signal Processing
• Machine Learning

Background:

• Electrocardiogram (ECG) signals offer a secure biometric modality due to inherent physiological traits, resisting forgery.
• Traditional biometric systems face challenges with spoofing and external attacks, necessitating more robust authentication methods.

Purpose of the Study:

• To develop a novel, real-time biometric authentication system leveraging Graph Convolutional Networks (GCN) and Mutual Information (MI) indices from ECG signals.
• To enhance the security and efficiency of biometric identification through advanced signal processing and machine learning techniques.

Main Methods:

• Extracted Mutual Information (MI) indices from 12-lead ECG signals to quantify statistical dependencies between leads.
• Constructed a graph representation using ECG features as nodes and MI values as edge weights.
• Trained a Graph Convolutional Network (GCN) model on the constructed graph for efficient user identification.

Main Results:

• The proposed GCN-MI model achieved 100% accuracy with a 5-layer architecture and a k-fold of 75.
• The system demonstrated superior performance compared to conventional methods, requiring less training data.
• The approach proved to be scalable and suitable for real-time applications.

Conclusions:

• The integration of MI indices and GCN provides a robust and efficient feature selection mechanism for ECG-based biometrics.
• The graph-based learning framework effectively captures spatial and statistical ECG data relationships, enhancing classification accuracy.
• This novel GCN-MI approach sets a new benchmark for real-time, secure biometric authentication in various applications.