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

The electrocardiogram (ECG) is a highly secure biometric modality due to its intrinsic physiological characteristics, making it resilient to forgery and external attacks. This study presents a novel real-time biometric authentication system integrating Graph Convolutional Networks (GCN) with Mutual Information (MI) indices extracted from 12-lead ECG signals.

The MI index quantifies the statistical dependencies among ECG leads and is computed using entropy-based estimations. This index is used to construct a graph representation, where nodes correspond to ECG features and edges reflect their relationships based on MI values. The GCN model is trained on this graph, enabling it to learn complex patterns for user identification efficiently.

Experimental results demonstrate that the proposed GCN-MI model achieves 100% accuracy with a 5-layer architecture at a k-fold of 75, outperforming conventional approaches that require less training data.

This work introduces several innovations: the integration of MI indices enhances feature selection, improving model robustness and efficiency; the graph-based learning framework effectively captures both spatial and statistical relationships within ECG data, leading to higher classification accuracy; the proposed approach offers a scalable and real-time biometric authentication system suitable for applications in finance, healthcare, and personal device access. These findings highlight the practical value of the GCN-MI approach, setting a new benchmark in ECG-based biometric identification.