Linkage Attacks Expose Identity Risks in Public ECG Data Sharing
Ziyu Wang, Elahe Khatibi, Farshad Firouzi, Sanaz Rahimi Mousavi, Krishnendu Chakrabarty, Amir M. Rahmani · 2025
Read Linkage Attacks Expose Identity Risks in Public ECG Data Sharing by Ziyu Wang, Elahe Khatibi, Farshad Firouzi, Sanaz Rahimi Mousavi, Krishnendu Chakrabarty, Amir M. Rahmani (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-15850.txt.
Introduction
Ziyu Wang and five co-authors show that shared heart signal data can betray who it came from. Their paper from twenty twenty-five studies electrocardiogram records posted for research. Heart traces act like fingerprints, which opens a linkage risk. Past privacy studies assumed foes with perfect knowledge. This team tests weaker, realistic foes with partial data.
Records from over a hundred people across real sets were used. Their method re-identified people in public sets with high accuracy at a tuned threshold. A share of strangers was also wrongly matched to known people. Plain anonymization failed to block the linkage. The authors call for stronger shields: differential privacy, access control, and encrypted computing. Such tools must guard identity while keeping the data useful for care research.
Catalog record
| CREATOR | Ziyu Wang, Elahe Khatibi, Farshad Firouzi, Sanaz Rahimi Mousavi, Krishnendu Chakrabarty, Amir M. Rahmani |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-15850
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Linkage Attacks Expose Identity Risks in Public ECG Data Sharing Ziyu Wang, Elahe Khatibi, Farshad Firouzi, Sanaz Rahimi Mousavi, Krishnendu Chakrabarty, Amir M. Rahmani Abstract The increasing availability of publicly shared electrocardiogram (ECG) data raises critical privacy concerns, as its biometric properties make individuals vulnerable to linkage attacks. Unlike prior studies that assume idealized adversarial capabilities, we evaluate ECG privacy risks under realistic conditions where attackers operate with partial knowledge. Using data from 109 participants across diverse real-world datasets, our approach achieves 85% accuracy in re-identifying individuals in public datasets while maintaining a 14.2% overall misclassification rate at an optimal confidence threshold, with 15.6% of unknown individuals misclassified as known and 12.8% of known individuals misclassified as unknown. These results highlight the inadequacy of simple anonymization techniques in preventing re-identification, demonstrating that even limited adversarial knowledge enables effective identity linkage. Our findings underscore the urgent need for privacy-preserving strategies, such as differential privacy, access control, and encrypted computation, to mitigate re-identification risks while ensuring the utility of shared biosignal data in healthcare applications. I Introduction Electrocardiograms (ECG) capture the heart’s electrical activity, serving as a key diagnostic tool for conditions like heart failure and arrhythmias [ 1 , 2 ] . Beyond clinical use, ECG signals exhibit unique morphological and temporal patterns influenced by heart anatomy, conduction pathways, and autonomic regulation, making them suitable for biometric identification [ 3 ] . Unlike fingerprints, ECG signals are dynamic and vary naturally due to physiological changes, enabling intrinsic condition detection, which reduces susceptibility to spoofing and enhances security in authentication systems [ 4 ] . As telehealth platforms and wearable devices integrate ECG data for remote monitoring [ 5 , 6 ] , the increasing availability of publicly shared datasets raises significant privacy concerns [ 7 ] . Due to their biometric nature, ECG signals are vulnerable to linkage attacks, where adversaries exploit overlaps between public and private datasets to re-identify individuals. Unlike demographic data, ECG retains distinctive patterns over time, making de-identification challenging. An attacker with partial knowledge can aggregate ECG samples from wearables, telehealth platforms, or leaked records and cross-reference them with public datasets using machine learning [ 8 ] . Leveraging membership inference [ 9 ] and record linkage techniques, attackers can link anonymized ECGs to identities, compromising patient privacy. This risk grows as ECG datasets proliferate across research, clinical, and commercial domains without adequate safeguards. Fig. 1: Illustration of a linkage attack on public anonymous ECG data. A telehealth platform integrates data from multiple sources, including wearable devices and medical institutions. An attacker with partial knowledge cross-references public and private datasets to re-identify individuals, compromising privacy and trust in medical data sharing. Figure 1 demonstrates this problem, showing how telehealth platforms consolidate ECG data from diverse sources, creating a rich repository that attackers can exploit [ 10 , 11 ] . By correlating leaked or externally obtained ECG signals with public datasets, adversaries can disclose identities, leading to privacy breaches. Beyond compromising confidentiality, linkage attacks undermine trust in medical data sharing, discourage research participation, and threaten the security of digital health infrastructures [ 12 ] . As healthcare increasingly relies on open data initiatives, mitigating these risks is essential to preserving patient privacy and trust [ 7 , 13 ] . Despite these risks, existing ECG privacy studies often make overly simplified assumptions about adversarial capabilities [ 14 ] , assuming full data access or controlled re-identification settings. In reality, attackers operate with partial and noisy knowledge, handling heterogeneous datasets from varying devices, medical conditions, and recording environments. Moreover, prior studies struggle to distinguish known from unknown individuals [ 3 , 15 ] , inflating success rates in controlled settings while overlooking real-world challenges like within-person variation due to aging, disease progression, or sensor differences [ 16 ] . These gaps hinder a true understanding of ECG privacy risks and the development of effective protective measures. To address ECG privacy risks, we propose the first framework to assess re-identification under realistic linkage attacks. Unlike prior studies assuming full adversarial knowledge, we model a practical threat where attackers with partial knowledge cross-reference auxiliary ECG data with public repositories to link identities, simulating real-world risks from wearables, telehealth, and leaked medical records. We implement a Vision Transformer (ViT)-based model [ 17 ] to assess how well attackers can match ECG signals while correctly distinguishing unknown individuals. Our study provides an accurate evaluation of linkage attack risks in biosignal data sharing, demonstrating the urgent need for stronger privacy-preserving measures in healthcare. Our contributions are as follows: • We define and formalize ECG linkage attacks, illustrating how adversaries exploit overlaps between public and private datasets for re-identification. • We introduce the first realistic attack model that quantifies re-identification risks under partial adversarial knowledge, aligning with real-world telehealth and data-sharing constraints. • We provide empirical evidence of ECG privacy vulnerabilities across diverse datasets, establishing benchmarks to guide privacy-preserving research. II Methodology Fig. 2: Overview of the proposed linkage attack method. The biometric nature of ECG signals makes publicly shared datasets vulnerable to linkage attacks, where adversaries exploit overlaps between public and private data to re-identify individuals. Figure 2 illustrates our ViT-based framework. First, raw ECG signals undergo pre-processing, segmentation, and embedding. A ViT encoder extracts identity-specific features, followed by an Multi-Layer Perceptron (MLP) classifier. Stage 1 (Identity Matching) leverages an adversarial dataset—collected from public or leaked sources—to train a ViT model that maps ECG signals to latent space representations, assigning identity probabilities via softmax. Stage 2 (Unknown Detection) applies a dynamic confidence threshold to reject uncertain matches, reducing false identifications. This framework quantifies re-identification risks, showing that adversaries can link ECG data even with partial knowledge, underscoring the need for stronger privacy safeguards. II-A Problem Definition Let \mathcal{D}_{\text{pub}}=\{(x_{i},y_{i})\}_{i=1}^{N} be a public ECG dataset, where x_{i}\in\mathbb{R}^{T} is an ECG signal of length T , and y_{i}\in\mathcal{Y} is the corresponding identity label. Similarly, let \mathcal{D}_{\text{priv}}=\{x_{j}\}_{j=1}^{M} be a private dataset containing ECG signals from both known participants , whose identities exist in \mathcal{D}_{\text{pub}} , and unknown participants , who are absent from \mathcal{D}_{\text{pub}} . Given x_{j}\in\mathcal{D}_{\text{priv}} , an attacker seeks to assign a label \hat{y}_{j}\in\mathcal{Y}\cup\{\text{unknown}\} , identifying whether the signal belongs to a known or unseen individual. We investigate linkage attack risks, where an adversary attempts to re-identify individuals by cross-referencing ECG samples in \mathcal{D}_{\text{priv}} with identities in \mathcal{D}_{\text{pub}} . The attacker uses an adversarial dataset—ECG signals collected from public sources, leaked medical data, or unauthorized means, containing identifiable records. Unlike standard classification, the attacker operates under partial knowledge, meaning they do not know whether a given identity exists in \mathcal{D}_{\text{pub}} . This introduces practical challenges, such as dataset mismatches, ECG variability across devices and conditions, and the need to differentiate between known and unknown individuals. Our study formalizes this threat model and quantifies the re-identification risks posed by such attacks in real-world data-sharing scenarios. II-B Framework for Linkage Attack Evaluation The proposed framework evaluates re-identification risks by simulating an attacker attempting to link ECG samples from \mathcal{D}_{\text{priv}} to individuals in a black-box public dataset \mathcal{D}_{\text{pub}} . Unlike traditional attacks that assume full adversarial knowledge [ 7 , 3 ] , our framework simulates real-world constraints, where the attacker has limited visibility and no prior knowledge