An ECG-based Authentication System Using Siamese Neural Networks

• Published in: Journal of medical and biological engineering Vol. 41; no. 4; pp. 558 - 570
• Main Authors: Ivanciu, Liliana, Ivanciu, Iustin-Alexandru, Farago, Paul, Roman, Marius, Hintea, Sorin
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.08.2021; Springer Nature B.V
• Subjects: Biological Techniques; Biomedical and Life Sciences; Biomedical Engineering and Bioengineering; Biomedical Engineering/Biotechnology; Biomedicine; EKG; Electrocardiography; Graphical representations; Neural networks; Original Article; Regenerative Medicine/Tissue Engineering; Rejection rate; Robustness (mathematics); Training; Biometrics; Electrocardiography; Feature extraction; Neural networks; Authentication; Cloud
• ISSN: 1609-0985; 2199-4757
• DOI: 10.1007/s40846-021-00637-9

Purpose Biometric systems are becoming increasingly important in today’s society. The Electrocardiogram signal proves a suitable contender for such systems thanks to its universality and robustness to attacks. We implement a cloud-based system for subject authentication using the Electrocardiogram signal and Siamese Neural Networks. Methods The key point of this approach consists in using images of the ECG signal, rather than numerical values, for training and deploying the model in our private cloud orchestrated by OpenStack. Results The experimental results were obtained using data from 90 subjects: the sensitivity of the authentication system is 87.3%, the False Rejection Rate is 12.7% and the False Acceptance Rate is 13.74%. The overall accuracy of the system is 86.47%. Conclusion This paper demonstrates the feasibility of an authentication system, deployed in a private cloud orchestrated by OpenStack, which uses Siamese Neural Networks and graphical representations of the ECG signal. Our contribution is two-fold: first, we make use of the inherent properties of Siamese Neural Networks to help simplify the training process and make it easier to enroll new subjects. Second, by deploying the model in our private cloud we not only ensure the portability, but also the scalability and security of the system.