Privacy-aware Early Detection of COVID-19 through Adversarial Training

Omid Rohanian, Samaneh Kouchaki, Andrew A. S. Soltan, Jenny Yang, Morteza Rohanian, David A. Clifton

Abstract

Early detection of COVID-19 is an ongoing area of research that can help with triage, monitoring and general health assessment of potential patients and may reduce operational strain on hospitals that cope with the coronavirus pandemic. Different machine learning techniques have been used in the literature to detect coronavirus using routine clinical data (blood tests, and vital signs). Data breaches and information leakage when using these models can bring reputational damage and cause legal issues for hospitals. In spite of this, protecting healthcare models against leakage of potentially sensitive information is an understudied research area. In this work, we examine two machine learning approaches, intended to predict a patient's COVID-19 status using routinely collected and readily available clinical data. We employ adversarial training to explore robust deep learning architectures that protect attributes related to demographic information about the patients. The two models we examine in this work are intended to preserve sensitive information against adversarial attacks and information leakage. In a series of experiments using datasets from the Oxford University Hospitals, Bedfordshire Hospitals NHS Foundation Trust, University Hospitals Birmingham NHS Foundation Trust, and Portsmouth Hospitals University NHS Trust we train and test two neural networks that predict PCR test results using information from basic laboratory blood tests, and vital signs performed on a patients' arrival to hospital. We assess the level of privacy each one of the models can provide and show the efficacy and robustness of our proposed architectures against a comparable baseline.

The Privacy Challenge in Healthcare AI

In the rush to deploy machine learning models for pandemic management, privacy often takes a backseat to performance. However, machine learning models trained on Electronic Health Records (EHR) are surprisingly good at "reading between the lines." Even if explicit demographic columns like "Gender" or "Ethnicity" are removed, neural networks can often reconstruct this information because clinical markers—such as hemoglobin levels or white blood cell counts—frequently correlate with demographic groups.

Figure 1: Threat model showing how an adversary attempts to extract sensitive information from a trained model or its API.

This creates a major privacy risk known as a "property inference attack." As illustrated in Figure 1, if an adversary gains access to a trained model or its API, they might extract sensitive demographic profiles of the training population. In a healthcare setting, this represents a significant legal and ethical liability. The challenge lies in building a model that is clinically effective for diagnosis but "blind" to sensitive traits it does not require.

Adversarial Training Methodology

To address this, researchers developed deep learning models that detect COVID-19 while actively "scrubbing" sensitive demographic information from the model’s internal representations. The primary innovation is an Adversarial Training with Gradient Reversal (ADV) architecture.

Figure 2: The adversarial training architecture featuring a shared representation h and multiple discriminators D(h) designed to protect sensitive attributes.

As shown in Figure 2, the model consists of a main classifier (predicting COVID-19 status) and several "discriminators" (the adversaries). These discriminators attempt to guess the patient's age, gender, or ethnicity from the model's internal hidden state $h$. During training, a Gradient Reversal Layer (GRL) is employed. In the forward pass, it allows data to flow normally. However, during backpropagation, it flips the sign of the gradients coming from the discriminators. This forces the model to optimize away from representations that help the discriminators, effectively "un-learning" demographic associations while preserving diagnostic utility.

Datasets and Validation

The study utilized real-world data from four UK National Health Service (NHS) trusts:

  • Oxford University Hospitals (OUH): The primary training and initial testing set (>115,000 instances).
  • External Validation: To ensure generalizability, the models were tested on data from University Hospitals Birmingham (UHB), Bedfordshire Hospitals (BH), and Portsmouth Hospitals University (PUH).

The focus remained on "routine" data—blood gases, vital signs, and standard laboratory blood tests—typically available within an hour of hospital arrival.

Key Results and Insights

The experiments yielded three critical findings:

  1. High Diagnostic Accuracy: The adversarial models achieved an AUC of ~0.85 to 0.86 for COVID-19 detection, matching the performance of standard, non-private baselines. This suggests that demographic bias is not necessary for accurate screening.
  2. Effective Privacy Protection: The ADV model significantly obscured sensitive data. An attacker's success rate in guessing gender from internal representations dropped from an AUC of 0.91 (standard model) to 0.45 (essentially random chance).
  3. Robustness and De-biasing: Models showed no significant performance drop when cross-tested across different demographics, proving they learned generalizable clinical features rather than demographic shortcuts.

Conclusion

This framework demonstrates that clinical accuracy and patient confidentiality are not mutually exclusive. By baking privacy directly into the learning process through adversarial training, healthcare providers can deploy robust diagnostic tools that are naturally resistant to information leakage and demographic bias. Future work will look to extend these techniques to multi-class demographic variables and unstructured data types like medical imaging.

Summary updated

Abstract

Early detection of COVID-19 is an ongoing area of research that can help with triage, monitoring and general health assessment of potential patients and may reduce operational strain on hospitals that cope with the coronavirus pandemic. Different machine learning techniques have been used in the literature to detect coronavirus using routine clinical data (blood tests, and vital signs). Data breaches and information leakage when using these models can bring reputational damage and cause legal issues for hospitals. In spite of this, protecting healthcare models against leakage of potentially sensitive information is an understudied research area. In this work, we examine two machine learning approaches, intended to predict a patient's COVID-19 status using routinely collected and readily available clinical data. We employ adversarial training to explore robust deep learning architectures that protect attributes related to demographic information about the patients. The two models we examine in this work are intended to preserve sensitive information against adversarial attacks and information leakage. In a series of experiments using datasets from the Oxford University Hospitals, Bedfordshire Hospitals NHS Foundation Trust, University Hospitals Birmingham NHS Foundation Trust, and Portsmouth Hospitals University NHS Trust we train and test two neural networks that predict PCR test results using information from basic laboratory blood tests, and vital signs performed on a patients' arrival to hospital. We assess the level of privacy each one of the models can provide and show the efficacy and robustness of our proposed architectures against a comparable baseline. One of our main contributions is that we specifically target the development of effective COVID-19 detection models with built-in mechanisms in order to selectively protect sensitive attributes against adversarial attacks.
BibTeX
@article{rohanian2023privacy,
  author = {Omid Rohanian and Samaneh Kouchaki and Andrew A. S. Soltan and Jenny Yang and Morteza Rohanian and Yang Yang and David A. Clifton},
  title = {{Privacy-aware Early Detection of COVID-19 through Adversarial Training}},
  journal = {IEEE J. Biomed. Health Informatics},
  year = {2023},
  url = {https://openreview.net/forum?id=q2V4oZusGe},
}
Figure