PEA: Parallel electrocardiogram-based authentication for smart healthcare systems

https://doi.org/10.1016/j.jnca.2018.05.007Get rights and content

Abstract

Currently, ECG-based authentication is considered highly promising in terms of user identification for smart healthcare systems because of its inimitability, suitability, accessibility and comfortability. However, it is a great challenge to improve the authentication accuracy, especially for scenarios that include a large number of users. Thus, this paper proposes a parallel ECG-based authentication called PEA. Specifically, this paper proposes a hybrid ECG feature extraction method that integrated fiducial- and non-fiducial-based features to extract more comprehensive ECG features and thereby improve the authentication stability. Furthermore, this paper proposes a parallel ECG pattern recognition framework to improve the recognition efficiency in multiple ECG feature spaces. Through the experiments, the performance of the proposed authentication is verified.

Introduction

To improve the degree of certain qualities or attributes, such as availability, privacy, reliability, safety, security, and their nonintelligent counterparts, smart healthcare systems focusing on these qualities are intended to improve health outcomes, reduce costs, and enhance the quality of life (Laplante et al., 2016; Gravina and Fortino, 2016). Because these systems involve various sensitive information, such as medical data and privacy records, it has become an indispensable task to safeguard the security and privacy of smart healthcare systems (Li et al., 2016). Therefore, various authentication methods have been proposed to protect user privacy and security.

It is now widely recognized that biometrics are more reliable than knowledge- and possession-based approaches such as identity cards and usernames/passwords because for biometrics, there is no need to remember anything. Biometric attributes cannot be lost, transferred or stolen, and they offer better security because these attributes are very difficult to forge and require the presence of a genuine user to grant access to particular resources (Unar et al., 2014). Therefore, the biometric-based approach is considered to play a critical role in balancing privacy with performance (Dantcheva et al., 2016). Generally, biometrics are divided into the following two categories:

  • 1.

    Behavioral Biometrics are often implemented on mobile and wearable devices to identify users by gesture, touch dynamics, keystroke, etc. (Abate et al.,; Bo et al., 2013). Generally, behavioral biometrics are used to prevent inside attacks, but they have also been deemed valid for entry into a system (Manning, 2017). Hence, some advanced biometrics involving more behavioral information are proposed to provide better authentication (Peng et al., 2017).

  • 2.

    Physiological Biometrics are powerful emerging modalities and are becoming a promising technology for automatic and accurate individual recognition in human identification, including electrocardiograms (ECGs) (Zhang et al.,) and electroencephalograms (EEGs) (Kumari and Vaish, 2015). For example, Martinovic, et al. proposed a pulse-response biometric system to enhance the security of continuous authentication on a secure terminal (Martinovic et al., 2017). More specifically, Barra, et al. proposed a physiological biometric system based on the extraction of fiducial features (peaks) from the ECG combined with the spectrum features of the EEG to support better authentication in healthcare applications (Barra et al., 2017).

  • 3.

    Multimodal Biometrics have been proposed to combine physiological and behavioral biometrics to improve the robustness (Bansal et al., 2017). In (Gowda et al., 2017), Gowda, et al. developed a hybrid biometric system in which both psychological and behavioral traits are fused at the score level, including face, palm, signature and speech traits. Furthermore, Sultana, et al. proposed mining social behavioral information from an online social network and fused traditional face and ear biometrics to enhance the performance of a traditional biometric system (Sultana et al., 2017).

Obviously, physiological biometrics are suitable for authentication in healthcare systems because the complexity and scalability that behavioral and multimodal biometrics often need to collect more data are unconnected with user healthcare. Therefore, ECG-based biometrics are widely used to provide continuous authentication for healthcare systems (Satija et al., 2017; Zebboudj et al., 2017; Zaghouani et al., 2017).

However, based on a comprehensive investigation of ECG-based authentication, in (Fratini et al., 2015), A. Fratini et al. concluded that that new techniques will be developed to improve the authentication accuracy, especially for scenarios that include a large number of users. Hence, this paper proposes a parallel ECG-based authentication named PEA for smart healthcare systems to provide more accurate and effective biometrics. Specifically, the main contributions of this work include the following:

  • Addressing the instable accuracy of authentication in different scenarios, this paper proposes a hybrid ECG features extraction integrating fiducial and non-fiducial based features. This approach attempts to extract more comprehensive ECG features to improve the authentication stability.

  • To improve efficiency, this paper proposes parallel ECG pattern recognition based on MapReduce that can effectively search the multimodal ECG feature space.

The remainder of this article is organized as follows. Section 2 presents the detailed design of the proposed scheme. Section 3 describes the proposed hybrid ECG feature extraction consisting of fiducial and non-fiducial features. The proposed parallel ECG pattern recognition is introduced in Section 4, followed by the experimental analysis in Section 5. Finally, Section 6 concludes this paper.

Section snippets

Motivation and design issues

The proposed ECG-based authentication is considered to be highly promising in terms of user identification for smart healthcare systems because of its attractive features:

  • Inimitability: Many significant works have proved that ECG is an inherent vital signal that cannot be easily imitated, unlike fingerprints, voice, iris and other biometrics (Arteaga-Falconi et al., 2016).

  • Suitability: ECG is more crucial than other physiological signals that some biometrics are not available for those who are

Hybrid ECG feature extraction

To accurately represent ECG features, as much information as possible should be involved in feature extraction. Therefore, this paper proposes a hybrid ECG feature extraction technique that includes fiducial- and non-fiducial-based features.

Theory model

Parallel processing is the key technique that support incremental training, in which only new data must be trained rather than re-training all of the data when new data are generated. Fig. 3 illustrates the detailed theory model of the proposed algorithm.

  • 1.

    Assisted by the distributed infrastructure of MapReduce, the raw data are divided into multiple subsets, Sample1, …, SampleN, for training and recognition.

  • 2.

    Each subset representing the hybrid features of the ECG signal is partitioned into

Experimental dataset and environment

To verify the performance of the proposed scheme, the MIT-BH database1 is considered the main experimental dataset. Specifically, MIT-BH includes 100 samples, and each sample contains 200 ECG signal files; thus, this dataset includes 20, 000 ECG signals. Moreover, 100 morbid samples are added to the experimental dataset by considering the diversity of the data sources. Each sample includes 10, 000 signals that are equal to the volume of the ECG signal of one user that is

Conclusions

To improve the accuracy and efficiency of ECG-based authentication, this paper proposed a parallel approach that incorporates multiple features for smart healthcare systems. Specifically, fiducial- and non-fiducial-based features, i.e., PQRST, spectral and morphological features, are comprehensively considered for ECG recognition. Assisted by a parallel computing framework, the recognition is divided into the following two processing methods: searching local optima in each feature space and

References (30)

  • A. Dantcheva et al.

    What else does your biometric data reveal? a survey on soft biometrics

    IEEE Trans. Inf. Forensics Secur.

    (2016)
  • A. Fratini et al.

    Individual identification via electrocardiogram analysis

    Biomed. Eng. Online

    (2015)
  • H.D.S. Gowda et al.

    Combination of physiological and behavioral modalities based on score level fusion

  • R. Gravina et al.

    Automatic methods for the detection of accelerative cardiac defense response

    IEEE Trans. Affect. Comput.

    (2016)
  • K. Huang et al.

    Semisupervised sparse multilinear discriminant analysis

    J. Comput. Sci. Technol.

    (2014)
  • Cited by (0)

    View full text