A Review on ECG based Human Authentication

Transcription

1 A Review on ECG based Human Authentication Pooja Ahuja 1, Abhishek Shrivastava 2 1 Dept of CSE, DIMAT,Raipur, India 2 Dept of CSE, DIMAT,Raipur, India Abstract- Biometric systems are mostly used for human identification and authentication. Recent developments have shown that for person identification ECG can be used as more powerful tool as it give more reliable and accurate results even in abnormal cases than other biometric characteristics. Credential based authentication methods (e.g., passwords, PINs, certificates) are not well-suited for remote healthcare as patients could give credentials to someone else. Furthermore, one-time authentication using trait-based biometrics (e.g., face, fingerprints, and iris) or credentials do not cover the entire monitoring period and may lead to unauthorized post-authentication use. In this paper various techniques previously proposed in literature for feature extraction and a comparative study to predict the accuracy of overall system is being discussed. Keywords-ECG; Feature extracted; ECG fiducial method; ECG non-fiducial method; Classification; identification; LDA (linear discriminant analysis); PCA (principal component analysis); WT (wavelet transformation) I. INTRODUCTION Biometric measures are among rapid-emergent fields of information security, gradationally entering into all areas of human activity. Importance of utilizing ECG as a biometric tool is supported by the fact that ECG is a universal characteristic, as the heart beat is a necessary for life, and it can be recorded with minimum inconvenience to the individual ECG analysis is not only a very useful tool for clinical diagnosis, but also is recently studied as a potential biometric.ecg is a method to measure and record different electrical potentials of the heart. Willem Einthoven developed the ECG method in the early 1900s [6]. ECG biometric features can be classified as either characteristic based features [7, 13] or waveform based feature [8,9]. Characteristic based features are easy to obtain, since to start the extraction process only the information of the ECG fiducial points in one ECG complex are required. Fiducial points are the points that correspond to the peaks and boundaries of the three major waves in an ECG signal namely the P, the QRS, and the T waves. There are 3 peak locations (P, R, T) and 6 boundaries in a typical ECG complex. Fig. 1 a sample ECG signal showing P-QRS-T All Rights Reserved 757

2 The ECG may be grouped into the phases of depolarization and repolarization of the muscles fiber making up the heart. The depolarization correspond to the P-Wave (atrial depolarization and QRS wave (ventricles depolarization). The repolarization corresponds to the T-wave. ECG (electrocardiogram) signal is one of the most well-known biomedical signals for abnormality recognition. ECG captures cardiac features from subject that are distinctive in nature. It is been recorded with electrodes attached at surface of body. The basic difference between ECG and other biometrics is that they are easy to impersonate and forge, as they are external biometrics and due to this an internal biometric is far more reliable.ecg is emerging as a mainly fascinating biometric trait in multi-biometrics scenarios, as it is: a) Originated in the body by a vital organ. b) More difficult to imitate. c) Permanently accessible, providing a continuous and near-ubiquitous means of recognition. Human identification based on ECG signal can be classified under two major category, fiducial methods and non fiducial methods. 1. Fiducial methods- It depends on local features of heart beats for biometric template design, such as temporal or amplitude difference between consecutive fiducial points and used for identification purposes. The drawback of fiducial features is their sensitivity to noise. Furthermore, detection of fiducial features in abnormal cases with arrhythmia may include errors in data. 2. Non fiducial method- It was initially proposed by Plataniotis et al. [13] in order to remove the necessity of fiducial point s localization of the ECG signal. It treat ECG signal or secluded heart beats holistically and captures features based on overall morphology of waveform [13]. II. LITERATURE SURVEY Although extensive research has been conducted for ECG based clinical applications, the research for ECG-based bio-metric identification is relevant. Biel et al. [1] are among the earliest effort that conducted experiment on 20 subjects ECG record and showed the possibility of using ECG for human identification purposes. A set of twelve temporal and amplitude features are extracted from SIEMENS ECG equipment directly. A multivariate analysis-based method is employed for classification. The major drawback of Biel et al work lies in extraction of the ECG features by specific equipment which results in lack of automatic recognition. Irvine et al. [2] introduced a system to utilize heart rate variability (HRV) as a biometric for human identification. Israel et al. [3] subsequently proposed a more extensive set of descriptors to characterize ECG trace under various stress conditions. Initially ECG signal is first preprocessed by a bandpass filter. ECG processed is followed is followed by logical series of experiments with quantifiable metrics. Data filters are designed and fiducial points were identified. A Wilks Lambda method is utilized for feature selection and linear discriminant analysis for classification. The experiment was conducted on a database of 29 subjects with around 81% heartbeat recognition rate and 100% human identification rate can be achieved. In a later work, to overcome the limitations of ECG signal Israel et al.[4] presented a multimodality system that combines face and ECG signal for biometric identification. Although Israel et al s method provides automatic recognition, but due to the insufficient representation of the feature extraction methods the identification accuracy with respect to heartbeat is low Shen et al. [5] introduced a new approach that utilizes techniques of template matching and neural networks. Firstly template matching method is utilized to calculate the correlation coefficient for comparison of two QRS complexes. Then, to complete the verification, a decision-based neural network (DBNN) is used. The experimental results from 20 individuals showed that the correct verification rate was 95% for template matching, 80% for the DBNN, and 100% after combining both the methods. Shen [6] extended the proposed framework in a larger database of 168 normal healthy subjects. Template matching and mean square error (MSE) methods were compared for prescreening, and DBNN and distance classification compared for second-level classification. Total seventeen temporal All Rights Reserved 758

3 amplitude features was employed for the second-level classification. The best identification rate for 168 subjects is 95.3% using template matching and distance classification. Fatemian et al. [7] proposed a new wavelet based framework for automatic analysis of ECG for human recognition. In these personalized heartbeat template. Is designed which reduces the storage requirement of system substantially. Experimental results for subject identification indicate a robust subject identification rate of 99.61% over PTB and MIT healthy ECG databases using only 2 heartbeats in average for each subject. Chan et al. [8] introduced another non-fiducial feature extraction approach using a set of distance measures. Data was extracted using button electrodes held between the thumb and finger. The accuracy from 50 subjects was 89% using wavelet transform nonfiducial feature extraction approach. But these studies did not consider the impact of activity that contributes to physiological fluctuation (such as increased individual s heart rate) and noise within the ECG signal spectrum.can Ye et al [9] use a two-lead ECG signals for human identification. Wavelet Transform and Independent Component Analysis are applied to individual lead signal to extract morphological information. This information from two leads is combined by rejecting the heartbeat segments. The classification is based on the majority voting among multiple consecutive consistently classified heartbeats. The accuracy rate is 99.6 % that illustrates the great potential of ECG signals and the proposed method in the biometrics system. Wang et al. [10] were the first to introduce an approach by combining a set of analytic features derived from Fiducial points with appearance features obtained using PCA and LDA(principal component analysis and linear discriminate analysis) for feature extraction and data reduction which is free from fiducial detection. The accuracy for 13 subjects was 84% using analytic features alone and 96% using LDA with K-NN (K-nearest neighbor). To achieve 100% accuracy combination of the types of features was used. B. Vuksanovic and M. Alhamdi [11] presented another feature extraction approach using AR modelling.in these identification has been attempted using extracted analytic features (amplitude, time and width) and modelling features (AR parameters). K nearest neighbour (knn) classification algorithm applied in order to classify those features and evaluate the proposed approach. Md. Khayrul Bashar [12] introduces a system that extracts multiscale geometric features from ECG signals and applies them for human identification. By using all extracted features (Amplitude, Time, Width and AR model) to identify individual, it is possible to reach 100% accuracy of classified subjects with a reasonable negative test value. III. BASIC STEPS INVOLVED IN ECG ECG biometric system is a pattern recognition system that operates by acquiring biometric data from an individual, extracting a feature set from the acquired data, and comparing this feature set against the template set in the database [19]. The steps involved in such a system are: 1) By the use of sensors, first the signal is acquired. 2) The signal is then preprocessed and described in a suitable representation. 3) Next feature extraction is done. 4) A classification block processes the features and delivers a decision corresponding to system. Figure2. Signal acquisition Preprocessing Feature extraction Classification Fig. 2 Basic steps while processing All Rights Reserved 759

4 3.1 Signal Acquired This is the first step in any recognition or authentication system. In this step signal is being acquired by help of sensors which is then stored in database and is further used by system for matching purposes. 3.2 Preprocessing of ECG The first step in the analysis of ECG signal is the removal of noise from ECG signal. Denoising or pre-processing of ECG signal is vital because noise severely limits the effectiveness of the recorded ECG.Due to presence of noise the feature extraction and classification becomes less accurate. For ultimate data structure, the raw ECG data must be processed. Therefore denoising process is important to minimize the negative effects of noise, for example Wang uses [18] butter worth bandpass filter to perform noise reduction. A thresholding method is then applied to eliminate the outliers that are not suitable for training and classification Based upon which a filter is designed and applied to the raw data. Then after filtering feature extraction is performed. The steps involved in the preprocessing phase are: Digital filtering: It is the main step of preprocessing. It is used to filter unnecessary noise in ECG signals. Example the digital notch filter is employed to remove the power line interference. A high pass band filter with a cut off frequency of 0.5 Hz is used to remove the DC offset on the acquired signal. Down Sampling: It is a consistent solution for algorithms performance on high sampled signals. Peak detection: In peak detection, the unprocessed signal is digitally filtered and the peak index detected is used in the feature selection setup. Segmentation: In this step particular peak of signal is identified first then ECG waveform is segmented into individual heartbeat. 3.3 Feature Extraction After pre-processing, the second stage towards classification is to detect certain features of ECG signals mostly QRS complex, P and T waves. The features, which represent the classification information contained in the signals, are used as inputs to the classifier used in the classification stage. The goal of the feature extraction stage is to find the smallest set of features that enables acceptable classification rates to be achieved. In general, the developer cannot estimate the performance of a set of features without training and testing the classification system. Therefore, a feature selection is an iterative process that involves training different feature sets until acceptable classification performance is achieved. The feature extraction is the main criteria for further processing. Statistical features like dower matrix, correlation coefficient, covariance matrix etc,wavelet features like ICA, Fourier transform, Discrete cosine transform (DCT) etc techniques are used to extract the features of ECG. Till now various research and methods had been proposed for feature extraction of ECG signal which is illustrated in following table All Rights Reserved 760

5 Table 1: Different feature extraction techniques from an ECG signal proposed by various researchers Papers Studied Biel[1] (2001) Methodology used 1. Uses SIEMENS ECG apparatus to record. 2. Select appropriate medical diagnostic features for classification. Feature Selection Methods Inspection of correlation Matrix Shen [6] (2002) 1. This paper uses one-lead ECG signal to identify person from group. 2. Use template matching and neural networks to classify QRS complex related characteristics. Template matching and neural network Palaniappan[17] (2004) Israel [3] (2005) Plataniotis[13] (2006) 1. It proposes the use of form factor of QRS segment. 2. Uses two different neural network architectures for classification of six QRS wave related features. 1. Analysis fiducial based temporal features under various stress conditions. 2. ECG processed is followed is followed by logical series of experiments with quantifiable metrics. 3. Data filters are designed and fiducial points were identified. 1. Analyze the autocorrelation of ECGs feature extraction and apply DCT for dimensionality reduction. 2. ECG biometric recognition method that doesn t require any waveform detection is introduced. Form Factor Wilk s Lamda Discrete Cosine Transform (DCT) Gahi[14] (2008) 1. In this paper system extract 24 temporal and amplitude features from ECG signal, which on processing are reduced to 9 most relevant features Information Gain Ratio (IGR) Wang[10] (2008) Fang[15] (2009) 1. In this ECG signal is segment based on localized of R peak. 1. It provides unsupervised ECG based identification method. 2. Based on phase space reconstruction of onelead or three-lead ECG. 3. Saving from picking up characteristic points PCA or LDA Portrait All Rights Reserved 761

6 Fatemian[7] (2009) 1. Less templates per subject in gallery set to speed up computation and reduce memory requirement. 2. Paper provides robust preprocessing stage that enables it to handle noise and outliers. 3. Design of personalized heartbeat template. Maximum Correlation Md Saiful Islam (2013) 1. A single threshold is used to determine QRSs from local minima of energy making the method linear i.e. O(n) in computational time which is O(n2) for slope based methods for a signal of n samples. First order derivative B. Vuksanovic and M. Alhamdi[11] (2013) 1. In this paper, identification has been attempted using extracted analytic features (amplitude, time and width) and modelling features (AR parameters) 2. k nearest neighbour (knn) classification algorithm applied in order to classify those features and evaluate the proposed approach AR modelling Md. Khayrul Bashar [12] [2015] 1. In this paper system extracts multiscale geometric features from ECG signals and apply them for human identification. Multiscale Pattern distribution 3.4 Classification after preprocessing and feature extraction The next step is to classify the input. The goal of classification is to identify a subject or to verify an identity claim from the sensor observations. This is done to check the robustness of system. There are various classification algorithms designed for this reason, namely Radial Basis Function (RBF), K Nearest Neighbor (knn), Bayes Network (BN), Multilayer Perceptron (MLP), and etc [17]. Bayes Network: BN is a directed acyclic graph (DAG) over a set of variables called U, where U = {x1... xk} and k 1. BN is a network structure represented as BS and shown as follows: BS = {p (u pa (u)) u U} (1) Where pa (u) is the set of parents of u in BS. Classification task begins by classifying y = x0 called the class variable given a set of ECG attributes, x = x1...xk [17] Multilayer Perceptron: MLP is a feed forward artificial neural network model with one or more layers between input and output layer in a directed graph. The input layer consists of the output of the proposed biometric sample extraction method one or more unseen layers and an output layer which find outs the subject s identity. But for the input nodes, each node consists of at least one neuron with a nonlinear activation function All Rights Reserved 762

7 Radial Basis Function: Radial Basis Function introduced as a variant of Artificial Neural Networks. Due to the non-linearity characteristics of RBF modelling of complex mappings can be done which requires multiple intermediary layers in perceptron architecture [16]. The error between the target and desired output is reduced using Gradient Descent Algorithm [16]. K Nearest Neighbor: KNN is an instance based learning algorithm. It defines hypothesis directly from ECG training examples and has the potential to adhere its model to previously unseen data. It searches for the most related element to a given query element with similarity defined by the standard Euclidean distance [17]. IV. CONCLUSION AND FUTURE WORK The electrocardiogram (ECG) is a noninvasive and the record of variation of the bio-potential signal of the human heartbeats. The ECG detection which shows the information of the heart condition is essential to improve the patient living quality and suitable cure. The extracted feature from the ECG signal plays a very important role in diagnosing the heart disease. This proposed paper also showed a comparative study evaluating the performance of different algorithms that were implemented before for ECG signal feature extraction. The future work essentially focuses on fast and correct feature extraction by utilizing more statistical data. Moreover our future work also has an eye on enhancement in diagnosing the heart disease. The essential parameters that must be considered are the accuracy of the algorithm and simplicity of the algorithm in providing the best results in feature extraction while developing an algorithm for feature extraction of an ECG signal V. REFERENCES 1. Biel, L, Pettersson O, Philipson L, Wide P,"ECG analysis: a new approach in human identification," Instrumentation and Measurement, IEEE Transactions on, vol.50, no.3, pp , Jun Irvine, J. M., Wiederhold, B. K., Gavshon, L. W., & et al "Heart rate variability: a new biometric for human identification", International Conference on Artificial Intelligence, pp , S. A. Israel, J. M. Irvine, A. Cheng, M. D. Wiederhold, Brenda K., Wiederhold, "ECG to identify individuals", Pattern Recognition, vol.38,, pp , Issue 1, Jan J. M. Irvin and S. A. Israel, "A Sequential Procedure for Individual Identity Verification Using ECG, EURASIP Journal on Advances in Signal Processing, Vol. 2009, 2009, Article ID: , pp Shen, T.W, Tompkins, W. J, Hu, Y.H,"One-lead ECG for identity verification," 2nd Conf. of the IEEE Eng. in Med. and Bio. Society and the Biomed. Eng. Society, vol.1, pp.62-63, Shen J., Bao S., Yang L., Li Y., "The PLR-DTW Method for ECG Based Biometric Identification". Proceeding of the 33rd Annual International Conference of the IEEE, Engineering in Medicine and Biology Society; Boston, MA, USA, pp , 3 September Fatemian, S.Z., Hatzinakos, D., "A new ECG feature extractor for biometric recognition," 16thInternational Conference on Digital Signal Processing, pp.1-6, July D. C. Chan, M. M. Hamdy, A. Badre, and V. Badee. "Wavelet distance measure for person identification using electrocardiograms", Transactions on Instrumentation and Measurement, IEEE, Volume 57, issue 2, Can Y.,Miguel C., B.V.K Vijaya k. "Investigation of human identification using two lead electrogram signals";, Proceedings of the 4th International Symposium on Applied Sciences in Biomedical and Communication Technologies,IEEE, Volume 50, T Y. Wang, F. Agrafioti, D. Hatzinakos, K. N. Plataniotis, Analysis of Human Electrocardiogram (ECG) for Biometric Recognition, EURASIP Journal on Advances in Signal Processing, article ID , B. Vuksanovic, M. Alhamdi, " Analysis of Human Electrocardiogram for Biometric Recognition Using Analytic and AR Modeling Extracted Parameters" International Journal of Information and Electronics Engineering, Vol. 4, No. 6, November Md. Khayrul Bashar, Yuji Ohta, and Hiroaki Yoshida, "ECG-based Biometric Authentication Using Mulscale Descriptors" ICIIBMS,: Signal Processing, Computer Networks and Telecommunications, Track1, Plataniotis, K.N., Hatzinakos, Lee, J.K.M., "ECG Biometric Recognition without Fiducial Detection," Biometric Consortium Conference, Biometrics Symposium: Special Session on Research at the, pp.1-6, Gahi, Lamrani, M., Zoglat, Guennoun, M., Kapralos, El-Khatib, "Biometric Identification System Based on Electrocardiogram Data," New Technologies, Mobility and Security, pp.1-5, Nov. All Rights Reserved 763

8 15. S.C Fang and H.L Chan; "Human identification by quantifying similarity and dissimilarity in electrocardiogram phase space", Pattern Recognition, vol.42, pp , Issue 9, Sept S. Haykin, "Neural networks: A comprehensive Foundation", 2nd ed., Prentice Hall, Palaniappan, R., Krishnan, S.M," Identifying individuals using ECG beats," International Conference on Signal Processing and Communications, pp , Dec Sara Zokaee, Karim Faez, "Human Identification Based on Electrocardiogram and palm print", International Journal of Electrical and Computer Engineering (IJECE) vol.2, no.2, pp. 261~266, April All Rights Reserved 764