Manal Tantawi

Fiducial feature reduction analysis for electrocardiogram (ECG) based biometric recognition

Although the electrocardiogram (ECG) has been a reliable diagnostic tool for decades, its deployment in the context of biometrics is relatively recent. Its robustness to falsification, the evidence it carries about aliveness and its rich feature space has rendered the deployment of ECG based biometrics an interesting prospect. The rich feature space contains fiducial based information such as characteristic peaks which reflect the underlying physiological properties of the heart. The principal goal of this study is to quantitatively evaluate the information content of the fiducial based feature set in terms of their effect on subject and heart beat classification accuracy (ECG data acquired from the PhysioNet ECG repository). To this end, a comprehensive set of fiducial based features was extracted from a collection of ECG records. This feature set was subsequently reduced using a variety of feature extraction/selection methods such as principle component analysis (PCA), linear discriminant analysis (LDA), information-gain ratio (IGR), and rough sets (in conjunction with the PASH algorithm). The performance of the reduced feature set was examined and the results evaluated with respect to the full feature set in terms of the overall classification accuracy and false (acceptance/rejection) ratios (FAR/FRR). The results of this study indicate that the PASH algorithm, deployed within the context of rough sets, reduced the dimensionality of the feature space maximally, while maintaining maximal classification accuracy.

A wavelet feature extraction method for electrocardiogram (ECG)-based biometric recognition

This paper proposes a discrete wavelet feature extraction method for an electrocardiogram (ECG)-based biometric system. In this method, the RR intervals are extracted and decomposed using discrete biorthogonal wavelet in wavelet coefficient structures. These structures are reduced by excluding the non-informative coefficients, and then, they are fed into a radial basis functions (RBF) neural network for classification. Moreover, the ability of using only the QT or QRS intervals instead of the RR intervals is also investigated. Finally, the results achieved by our method outperformed the auto-correlation (AC)/discrete cosine transform (DCT) method where the DCT coefficients are derived from the AC of ECG segments and fed into the RBF network for classification. The conducted experiments were validated using four Physionet databases. Critical issues like stability overtime, the ability to reject impostors, scalability and generalization to other datasets have also been addressed.

An evaluation of the generalisability and applicability of the PhysioNet electrocardiogram (ECG) repository as test cases for ECG-based biometrics

The PhysioNet is a very popular internet-based ECG repository which provides open access to a variety of ECG datasets. The data is collected from subjects within a medical framework, with the intention of acquiring clinically relevant information from patients. Because of the convenience afforded by the internet, literally thousands of ECG records can be downloaded and used for non-medical purposes, such as biometrics. The purpose of this study was to evaluate the applicability and/or suitability of the PhysioNet ECG data for deployment within biometrics. The needs and mindset of a clinician may be quite different from that of a security engineer. This paper therefore attempts to provide a preliminary examination of the PhysioNet ECG data repository along these dimensions, emphasising the need to create methodologies in the context of biometrics that not only take these considerations into account, but integrates them into the biometric methodology.

A novel feature set for deployment in ECG based biometrics

In the last two decades, the Electrocardiogram (ECG) was introduced as a powerful biometric tool for personal identification. The vast majority of publications in the ECG based biometrics domain have focused on extracting fiducial based features for use in the identification task. Fiducial based features refer to the landscape of an ECG, which encompasses three complex waves for each heartbeat. The fiducial based approach requires calculating amplitude and temporal distances between 11 fiducial points that represent the peaks, valleys, onsets and offsets of these waves. The purpose of this research is to investigate the efficiency of a subset of 23 fiducial features that has the advantage of relaxing the detection process to include only five points that represent the peaks and valleys of the three complexes. For comparison, a super set of 36 fiducial features and the subset of 23 features were examined using radial basis functions (RBF) neural network classifier. A dataset of 35 records of 13 subjects from PTB Physionet database was used for training and testing purposes. Thereafter, the generalization ability of the system to other datasets was tested using another set of 38 subjects from PTB database. The results show the ability of the proposed subset of 23 features to maintain the identification accuracy and provide better generalization results than the super set.

Electrocardiogram (ECG): A New Burgeoning Utility for Biometric Recognition

Recently, Electrocardiogram (ECG) has been emerged as a new biometric trait. ECG as a biological signal has the advantage of being an aliveness indicator. Moreover, it is difficult to be spoofed or falsified. In this chapter, a comprehensive survey on the employment of ECG in biometric systems is provided. An overview of the ECG, its benefits and challenges, followed by a series of case studies are presented. Based on the survey, ECG based biometric systems can be fiducial or non-fiducial according to the utilized features. Most of the non-fiducial approaches relax the challenging fiducial detection process to include only the R peak yielding to more reliable features. However, the drawback of such approaches is that they usually resulted in high dimension feature space. Hence, a non-fiducial ECG biometric system based on decomposing the RR cycles in wavelet coefficient structures using discrete biorthogonal wavelet transform is introduced. These structures were reduced through a proposed two-phase reduction process. The first phase globally evaluates the different parts of the wavelet structure (five details and one approximation parts) and maintains those parts that preserve the system performance. However, the second phase excludes more coefficients by locally evaluating the coefficients of each part based on an information gain criterion. Our experiments were carried out with four Physionet datasets using Radial basis functions (RBF) neural network classifier. Critical issues like stability over time, ability to reject impostors and generalization to other datasets have been addressed. The results indicated that with only 35 % of the derived coefficients the system performance not only can be preserved, but also it can be improved.

Fiducial Based Approach to ECG Biometrics Using Limited Fiducial Points

The majority of electrocardiogram (ECG) based biometric systems utilize fiducial based features, derived from 11 landmarks (three peaks, two valleys and six onsets and offsets) detected from each ECG heartbeat. The onsets & offsets landmarks may be obscured by a variety of noise sources. Hence, sophisticated algorithms are usually needed for the detection of these points, which in turn increase computational load and also the results may be suboptimal. This work proposes the utilization of a reduced set of 23 features named ’PV set’, which only requires the detection of the five major peaks/valleys instead of all the11 landmarks. The performance of the ’PV set’ is evaluated in comparison with a super set of 36 fiducial features (including PV set) that based on all the 11 landmarks, in addition to IG and RS sets which are subsets of the superset selected based on Rough sets (RS)and information gain (IG) criterion respectively. The evaluation was drawn based on measuring quantities, such as subject identification (SI) accuracy, heartbeat recognition (HR) accuracy and receiver operating characteristic (ROC) curves. The proposed PV set achieved comparable results to the other sets and better results at high noise levels, yielding a reliable and computationally cheaper solution.

QT correction for fiducial ECG features based biometric systems

In this paper, an electrocardiogram (ECG) based biometric system is proposed. A QT correction step is introduced to obviate the impact of heart rate variability, instead of just normalizing the features by the corresponding RR duration. Consequently, both approaches were examined in this work. Two sets of fiducial features were investigated: a super set of 36 features and a reduced version of it. Radial basis functions neural network is used as a classifier. The evaluation of the system was performed on the basis of subject identification (SI) accuracy and heartbeat recognition (HR) accuracy. The experiments were conducted using a 50-subject database and the results revealed the superiority of the QT correction approach, especially over time.

Comparing Multi-class Approaches for Motor Imagery Using Renyi Entropy

One of the main problems that face Motor Imagery-based system is addressing multi-class problem. Various approaches have been used to tackle this problem. Most of these approaches tend to divide multi-class problem into binary sub problems. This study aims to address the multi-class problem by comparing five multi-class approaches; One-vs-One (OVO), One-vs-Rest (OVR), Divide & Conquer (DC), Binary Hierarchy (BH), and Multi-class approaches. Renyi entropy was examined for feature extraction. Three linear classifiers were used to implement these five-approaches: Support Vector Machine (SVM), Multinomial Logistic Regression (MLR) and Linear Discriminant Analysis (LDA). These approaches were compared according to their performance and time consumption. The comparative results show that, Renyi entropy demonstrated its robustness not only as a feature extraction technique but also as a powerful dimension reduction technique, for multi-class problem. In addition, LDA proved to be the best classifier for almost all approaches with minimum execution time.

Diagnosing Heart Diseases Using Morphological and Dynamic Features of Electrocardiogram (ECG)

In this paper, an automatic method is proposed for the heart beat classification of 15 classes mapped into five main categories. The proposed method is applied separately to both leads 1 and 2. Dynamic segmentation is considered to reduce the effect of the heart beat rate variation. The segmented beats are subjected to discrete wavelet decomposition (DWT) to extract the morphological features besides the dynamic features represented by four RR intervals. Principle component analysis (PCA) is considered to reduce the dimension of the extracted morphological features. After that, the reduced features are concatenated with the dynamic features and fed into Support vector machine (SVM) classifier. Finally, the rejection fusion step is applied to combine the results from both leads 1 and 2 with a 93.84% average accuracy and 99.5% overall accuracy having been achieved using MIT-BIH dataset as a validation database.

A Hybrid Hierarchical Method for Electrocardiogram (ECG) Heartbeat Classification

This paper proposes an automatic reliable two-stage hybrid hierarchical method for ECG heartbeat classification. The heartbeats are segmented dynamically to avoid the consequences of the heart rate variability. Discrete Wavelet Transform (DWT) is utilized to extract morphological features. The extracted features are then reduced by using Principle Component Analysis (PCA). Subsequently, the resulted features along with four RR features are fed into Support Vector Machine (SVM) to classify five categories. Thereafter, the heartbeats are further classified to one of the classes belonging to the assigned category. Two different strategies for classification have been investigated: One versus All and One versus One. The proposed method has been applied on data from lead 1 and lead 2. A new fusion step is introduced, where stacked generalisation algorithm is applied and different types of classifiers have been examined. Experiments have been carried out using a MIT_BIH database. The best overall and average accuracies obtained by the first stage are 98.40% and 97.50% respectively. For the second stage, 94.94% and 93.19% are the best overall and average accuracies obtained respectively. The best results are achieved using SVM with one versus one classification strategy for both stages and decision trees classifier for the fusion step.