이한주, 최석환, 김광남,

Vol. 35, No. 1, pp. 87-97, 2월. 2025
10.13089/JKIISC.2025.35.1.87, Full Text:
Keywords: De-Identification, Electrocardiogram, Membership Inference Attack, Artificial intelligence
Abstract

As the importance of information security has become prominent, biometric technologies have rapidly emerged as a key for user identity verification. Especially, electrocardiogram (ECG) signals are used for various purposes in many applications because they are difficult to falsify and provide high accuracy. Moreover, ECG signal analysis technology which analyzes ECG signals using artificial intelligence models has recently been emerging. However, in such scenarios, ECG signal data can be vulnerable to specific security threats like membership inference attack. In this paper, we proposed de-identification methods for ECG signal data to enhance robustness against membership inference attacks. Specifically, the proposed de-identification method uses random rounding, Gaussian noise, impulse noise, and sinusoidal noise to de-identify ECG signal data. From the experimental results using the MIT-BIH Arrythmia Database, it is observed that the proposed de-identification method provides robustness against membership inference attacks while minimizing the decrease in accuracy of ECG signal classification models.

Statistics
Cite this article