ECG-ATK-GAN: Robustness against Adversarial Attacks on ECGs using Conditional Generative Adversarial Networks

TL;DR

This paper introduces ECG-ATK-GAN, a novel conditional GAN architecture designed to improve the robustness of ECG-based arrhythmia classification systems against adversarial attacks, maintaining high accuracy under various attack scenarios.

Contribution

The paper presents the first conditional GAN specifically designed to defend against adversarial attacks on ECG signals, incorporating a class-weighted objective and new network blocks for better detection and classification.

Findings

Outperforms existing models on robustness against six attack types

Maintains high accuracy under both white-box and black-box attacks

Effective on two publicly available ECG datasets

Abstract

Automating arrhythmia detection from ECG requires a robust and trusted system that retains high accuracy under electrical disturbances. Many machine learning approaches have reached human-level performance in classifying arrhythmia from ECGs. However, these architectures are vulnerable to adversarial attacks, which can misclassify ECG signals by decreasing the model's accuracy. Adversarial attacks are small crafted perturbations injected in the original data which manifest the out-of-distribution shifts in signal to misclassify the correct class. Thus, security concerns arise for false hospitalization and insurance fraud abusing these perturbations. To mitigate this problem, we introduce the first novel Conditional Generative Adversarial Network (GAN), robust against adversarial attacked ECG signals and retaining high accuracy. Our architecture integrates a new class-weighted objective function for adversarial perturbation identification and new blocks for discerning and combining out-of-distribution shifts in signals in the learning process for accurately classifying various arrhythmia types. Furthermore, we benchmark our architecture on six different white and black-box attacks and compare them with other recently proposed arrhythmia classification models on two publicly available ECG arrhythmia datasets. The experiment confirms that our model is more robust against such adversarial attacks for classifying arrhythmia with high accuracy.