Abstract
Electrocardiogram (ECG) signals, while fluctuating with physical activity and time, possess intrinsic measurements suitable for biometric authentication. ECG is particularly promising due to its "liveness" requirement and resistance to spoofing. However, most existing algorithms struggle with non-healthy subjects. This paper presents the ECOIL (ECG Cycle Normalization and Interpolation) algorithm, designed to convert ECG signals into standardized template cycles regardless of sampling rates or cardiovascular health status. The proposed method achieves 100% accuracy for healthy individuals and 90.11% for cardiovascular disease (CVD) patients.
Introduction
Traditional biometrics like fingerprints and facial recognition are vulnerable to duplication. In contrast, the ECG signal is an internal physiological process that is difficult to fake. Despite its potential, ECG biometrics face two primary challenges: physiological variability (heart rate changes due to stress or activity) and pathological variability (distortions caused by arrhythmias or other heart conditions). Most current systems fail when applied to the millions of individuals with cardiovascular diseases.
The ECOIL Methodology
The researchers developed the ECOIL pipeline to transform raw ECG data into a consistent Template ECG Cycle (TEC).
- Preprocessing and Segmentation: Signals are denoised using a Butterworth Band-Pass filter. The So-Chan method identifies R-waves to segment the continuous signal into individual heart cycles.
- Interpolation (ECI): To account for varying hardware sampling rates, cubic spline interpolation is used to ensure every heartbeat contains an identical number of data points.
- Normalization (ECN): Both time and voltage are linearly scaled to a [0, 1] range. This step is crucial as it removes the influence of heart rate and signal amplitude, focusing on the underlying morphology.
- Template Generation: Using k-means clustering, the algorithm identifies the most representative cycles for an individual, ultimately calculating the median of the 10 most central cycles to create a robust template.
Experimental Evaluation
The study utilized the PTB Diagnostic ECG Database (52 healthy subjects) and the QT Database (105 subjects, including 91 CVD patients). Authentication was performed by calculating the Euclidean Distance between a new sample and the stored template.
Key Results
- Healthy Cohort: The system achieved a perfect 100% accuracy rate.
- CVD Cohort: Accuracy reached 90.11%, a significant improvement over traditional wavelet-based methods that often fail on pathological signals.
- Mixed Population: In a combined test of 157 subjects, accuracy remained high at 94.27%.
- Signal Duration: Performance improved with longer recordings, moving from 94.23% at 5 seconds to 98.07% at 20 seconds.
Discussion and Conclusion
The core innovation of ECOIL is its ability to treat pathological distortions not as noise, but as part of a unique "pathological signature" for biometric identification. By normalizing the signal, the system becomes hardware-independent and inclusive of individuals with heart conditions.
While highly effective, the system encountered difficulties with a small subset of CVD patients where extreme wave shifts (P and T waves) increased Euclidean distances beyond authentication thresholds. Future research will focus on larger, more diverse populations and refining the algorithm to handle the most complex cardiac pathologies.
Summary updated