T. Lugovaya
Department of Applied Mathematics and Computer Science, Electrotechnical University "LETI", Saint-Petersburg, Russian Federation

Abstract

This research investigates the feasibility of using the electrocardiogram (ECG) as a new biometric for human identification. It is well known that the shapes of the ECG waveforms depend on human heart anatomic features and are different for different persons. But it is unclear whether such differences can be used to identify different individuals. This research demonstrates that it is possible to identify a specific person in a predetermined group using a one-lead ECG. A one-lead ECG is a one-dimensional, low-frequency signal that can be recorded from electrodes on the hands. ECG fragments containing QRS complexes, P and T waves extracted from the ECG are processed by principal component analysis and classified using linear discriminant analysis. Using this method on a predetermined group of 90 subjects, the experimental results showed that the rate of correct identification was 96%.

Biometric technologies are among fast-developing fields of information security, gradually entering into all spheres of human activity. Today only three biometric methods have proved their efficiency, namely, identification based on fingerprints, iris or retina, and face. Hand geometry, voice, writing and typing dynamics, etc. are also useful, depending on the purpose and range of application.

This research aims to develop an identification system based on ECG (figure 1). ECG is assumed as an almost unique human characteristic because morphology and amplitudes of registered cardiac complexes are governed by multiple individual factors, in particular, by formation and position of the heart, presence and nature of pathologies, etc.

Figure 1. Example of ECG with standard notations.

The system feasibility was discussed in [1, 2]. This study suggests other data interpretation and classification techniques, with the system tested on a higher level of live input data. The respective findings are compared below (table 2).

The identification system uses a classical scheme including data preprocessing, formation of input data space, transition to reduced feature space, ECG cycle classification and ECG record identification.

For usability, it is necessary to be able to collect the ECG easily and quickly. The data collected for this study comprise the ECG-ID Database, consisting of 320 single-lead ECG recordings from 90 subjects, each 20 seconds long, sampled at 500 Hz with 12-bit precision. Since single-lead ECGs vary significantly within an individual depending on the lead (the locations of the electrodes used to observe the ECG), the choice of lead is important. Lead I is the potential difference between the left and right hands (LA - RA). It was chosen because it is easily measured and it is not sensitive to minor variations in electrode locations.

The raw ECG is often rather noisy and contains distortions of various origins. Both high and low frequency noise components are present (figure 3).

A. ECG with power-line noise	B. ECG with high-frequency noise
C. ECG with both power-line and high-frequency noise	D. ECG with baseline drift

Figure 2. Examples of noisy ECGs.

Frequency-selective signal filtering was implemented using a set of adaptive bandstop and low-pass filters (figure 3).

Figure 3. Examples of frequency-selective signal filtering results.

Baseline drift correction was implemented using multilevel one-dimensional wavelet analysis. The original signal was decomposed at level 9 using biorthogonal wavelets. The signal reconstructed using final approximation coefficients is assumed to be the drifting baseline, which is subtracted from the original signal (figure 4). This method shows good results in both cases of clear and rather noisy ECG signals.

Figure 4. Examples of baseline drift correction results.

This section focuses on initial feature space formation. Obviously information on cardiac performance is basically held in the cardiac cycle fragment containing the QRS complex and P- and T-waves (referred to here as the PQRST-fragment). Therefore the process begins with extraction of a set of R-peak synchronized PQRST-fragments (figure 5). The PQRST-fragment length is fixed at 0.5 sec (250 samples).

Figure 5. R-peak detection and PQRST-fragment extraction.

PQRST fragments are used as informative features. Therefore extracted PQRST fragments are processed to enhance their similarity as follows:

1.	Correcting PQRST fragment mutual "vertical" shift due to residual baseline drift:
2.	Culling distorted (due to breathing or motion artifacts) and pathological PQRST-fragments:
3.	Correcting PQRST-fragments depending on heart rate using Bazett's formula:

Thus in the initial feature space (dimension N=250) the ECG appears as a set of PQRST-fragments with each seen as a separate pattern at subsequent system stages, to be interpreted and classified independently.

The feature space is reduced using Principal Component Analysis (PCA) so that its dimension can be reduced to 30 (with the Kaiser criterion) or even to 10 (with the scree test).

Alternately, the feature space can be reduced with a wavelet transform (WT) providing the same space reduction but with slightly poorer final identification.

The resulting PQRST-fragment patterns are then classified in reduced feature space using linear discriminant analysis.

At the final stage, the ECG record identification is based on PQRST-fragment classification results.

Experimental studies involved 90 human volunteers. ECG records were made in the sitting position. Heart rate and physical and emotional state were not limited. The collected data set contains 320 records, of which 200 records were assigned to the training set and 120 records to the test set. Differentiation between the training and test sets aimed to provide for maximum performance complexity, i.e., maximum difference between records in different sets both in monitoring time and human physical state.

Averaged results of series of experiments on PQRST-fragment classification and ECG record identification with different feature space reduction methods are tabulated in Table 1. As shown there, the best results were obtained using 30 principal components, yielding 96% correct ECG identification in the test set.

Additionally, results from [1, 2] and this research are compared in Table 2.

As a result of this research a recognition system was developed to solve the problem of biometric human identification based on ECG on a sufficiently large set of input data. The findings represent primary arguments for using ECG as a biometric characteristic in various biometric access control problems, opening up a brand new perspective for the study of biometric technologies, with potential applications in security and modern amenity systems.

[1] Biel L, Pettersson O, Philipson L, Wide P. ECG analysis: a new approach in human identification. IEEE Transactions on Instrumentation and Measurement 2001 June; 50(3):808-812.

[2] Yi WJ, Park KS, Jeong DU. Personal identification from ECG measured without body surface electrodes using probabilistic neural networks. Proc 2003 World Congress on Medical Physics and Biomedical Engineering, Sydney, Australia, 2003 August.