ECG Biometrics Using Deep Learning and Relative Score Threshold Classification.

Area of Science:

• Computer Science
• Biometrics
• Machine Learning

Background:

• Biometrics utilizes pattern recognition for individual identification.
• Electrocardiograms (ECG) offer unique, difficult-to-reproduce traits for secure biometric applications.
• Intra-variability in ECG signals presents challenges for traditional biometric systems.

Purpose of the Study:

• To propose and evaluate two novel deep neural network (DNN) architectures, TCNN and RNN, for improved ECG-based biometrics.
• To address the intra-variability challenges in ECG signals for more robust identification and authentication.
• To compare the performance of TCNN and RNN against state-of-the-art methods.

Main Methods:

• Developed Temporal Convolutional Neural Network (TCNN) and Recurrent Neural Network (RNN) architectures.
• Integrated a Relative Score Threshold Classifier (RSTC) with both DNN models.
• Trained and validated models on public ECG databases: Fantasia, MIT-BIH, and CYBHi.

Main Results:

• TCNN achieved high accuracy (up to 100%) in identification and low Equal Error Rate (EER, as low as 0.0%) in authentication.
• RNN also demonstrated strong performance, outperforming previous state-of-the-art methods.
• Both proposed architectures significantly improved upon existing biometric recognition techniques.

Conclusions:

• TCNN and RNN architectures show superior performance for ECG-based biometrics.
• These deep learning approaches effectively mitigate ECG intra-variability.
• Future work can enhance robustness and reduce validation time through data enrichment and transfer learning.