REACT: REinforcement Learning-Based Adaptive ECG Anonymization and Privacy Threat Mitigation^*

Electrocardiogram (ECG) signals, as sensitive biometric data, are highly susceptible to re-identification attacks, presenting significant privacy risks in healthcare data sharing. Traditional anonymization techniques, such as generative models and heuristic perturbations, often fail to achieve an optimal balance between privacy protection and clinical utility. In this paper, we propose a Deep Reinforcement Learning (DRL)-driven framework for dynamic noise injection into ECG datasets. The framework utilizes a multi-objective reward function that aims to minimize adversarial attack effectiveness while preserving diagnostic accuracy. By adapting to various attack scenarios, our method ensures robustness against different adversarial models. Experimental results on the MIT-BIH dataset demonstrate that our approach significantly reduces attacker accuracy to near-random levels, while maintaining high clinical performance. Further analyses, including Pareto-front analysis, entropy, and AUC scores, confirm that our method surpasses conventional noise injection techniques, achieving an improved trade-off between privacy preservation and clinical utility.