ECG-based Biometric Recognition without QRS Segmentation: A Deep Learning-Based Approach.

Area of Science:

• Biometrics
• Cardiovascular Signal Processing
• Machine Learning

Background:

• Electrocardiogram (ECG)-based identification systems typically require R-peak detection and beat averaging for feature extraction.
• These preprocessing steps introduce delays and can be influenced by inter-beat variations, impacting identification accuracy.
• Existing systems are also vulnerable to unregistered subjects.

Purpose of the Study:

• To propose a deep learning-based ECG biometric identification scheme that eliminates the need for R-peak detection and beat averaging.
• To address the vulnerability of identification systems to unregistered subjects.
• To achieve high identification accuracy with reduced processing time.

Main Methods:

• A deep learning model was developed for ECG biometric identification.
• The proposed method utilizes random ECG segments, bypassing traditional R-peak detection and beat averaging.
• The system was designed to handle unregistered subjects.

Main Results:

• An identification rate of 99.1% was achieved in a system with 235 enrollees.
• An equal error rate (EER) of 8.08% was recorded.
• The method demonstrated effectiveness in identifying individuals from random ECG segments.

Conclusions:

• The proposed deep learning approach offers a faster and more robust ECG biometric identification system.
• Eliminating R-peak detection and beat averaging significantly reduces identification delay.
• The scheme effectively addresses the challenge of unregistered subjects in biometric identification.