ECG Identification Based on the Gramian Angular Field and Tested with Individuals in Resting and Activity States

Area of Science:

• Biometrics
• Signal Processing
• Machine Learning

Background:

• Biosignals, particularly electrocardiograms (ECGs), are increasingly used in biometrics.
• Wearable devices enable reliable ECG acquisition, advancing real-world biometric applications.
• Existing ECG identification methods have limitations that necessitate novel approaches.

Purpose of the Study:

• To propose and validate a novel ECG identification system.
• To explore the application of Gramian Angular Fields (GAF) for ECG signal transformation.
• To assess the performance of a VGG19 convolutional neural network for ECG-based biometrics.

Main Methods:

• ECG recordings were transformed into Gramian Angular Field (GAF) images.
• A VGG19 convolutional neural network was tuned for the identification task.
• Experiments were conducted on two public datasets: MIT-BIH Normal Sinus Rhythm Database and ECG-GUDB.
• Performance was evaluated using accuracy, False Acceptance Rate (FAR), and False Rejection Rate (FRR).

Main Results:

• The proposed ECG identification system achieved an accuracy of 91% on both datasets.
• The system demonstrated a False Acceptance Rate (FAR) eight times higher than the False Rejection Rate (FRR).
• The feasibility of using GAF images for ECG identification was confirmed.

Conclusions:

• Transforming ECG signals into GAF images is a viable technique for biometric identification.
• The VGG19 convolutional neural network effectively identifies individuals using GAF-encoded ECG data.
• This approach offers a promising direction for developing advanced wearable biometric systems.