Prioritizing Patient Privacy

Big data and big privacy typically don’t go hand-in-hand, but designing healthcare technologies with privacy in mind could be the key to their widespread adoption.

AsianScientist (Sep. 30, 2021) – When COVID-19 infection rates skyrocketed around the world, medical frontliners found themselves overwhelmed by recurring waves of infections, hitting over half a million new cases globally each day in late 2020. Throughout the year, nightmarish tales of crowded hospital beds and dwindling oxygen supplies characterized healthcare systems on the brink of collapse.

As hospitals became the pandemic’s battleground, patients and doctors turned to digital health systems like teleconsultations to shift some of the burden from physical centers. Electronic health records, combined with artificial intelligence (AI) programs for analyzing patient data, enabled clinicians to automate disease diagnosis, predict recovery trajectories and deliver appropriate medical services.

While these data-driven technologies evidently enhance healthcare processes, concerns over privacy remain—raising important questions about ensuring the protection of personal information.


The boon and bane of big data

By providing increased accuracy and adaptability, powerful AI technologies have the potential to revolutionize healthcare, rapidly scouring masses of data to extract meaningful insights that evade even doctors’ trained eyes.

But as the name suggests, these big data applications work best when they can analyze vast amounts of clinical data. For example, to train an AI model to detect lung disease from chest computed tomography (CT) scans, it would need to see thousands of medical images and learn what healthy versus damaged cases typically look like. Though data may be the currency for these technologies, the full extent of their analytical power has yet to be harnessed.

After all, valid worries over data privacy tend to limit the amount and types of data that can be processed. Consider how personal preferences are already frequently used to curate content, such as in Netflix’s recommendation systems or tailored Facebook ads about a product a friend just mentioned. While these experiences are commonplace now, it doesn’t make them any less uncanny, almost as if the machine knows you a bit too well.

Healthcare data is especially sensitive—more so than other forms of data like climate patterns or economic trends—and contains personal information that many people would likely want to keep confidential. Without well-established data sharing regulations, there’s a real danger that these technologies can be used in intrusive, potentially harmful ways, providing a lopsided amount of power to those who control the information flow.


Placing privacy first

Given its magnitude and reach, the COVID-19 pandemic highlighted the need for big data healthcare applications to combat its spread and mitigate its disastrous effects. Accordingly, AI-based healthcare technologies are now being designed with built-in measures to protect patient privacy.

Today, researchers in Asia are developing privacy-preserving tech, overcoming the usual trade-offs between data security and accurate analytics. For example, a team at the Chinese University of Hong Kong (CUHK) recently built a high accuracy AI model for diagnosing COVID-19 from chest CT scans, without explicitly sharing the clinical data used.

“In the rapidly evolving pandemic of COVID-19, there has clearly been no time to set up complicated data sharing agreements across institutions, let alone countries,” shared lead author Assistant Professor Dou Qi from CUHK’s Department of Computer Science and Engineering.

To better understand clinical conditions while preserving data privacy, the researchers used a federated learning (FL) system, where the training and analysis processes are distributed to separate databases. In this decentralized framework, sensitive patient data is kept securely in a local model for each participating hospital, instead of feeding the data into a larger network accessible to others.

AI then uses the input data to learn the rules about how to perform the analysis most efficiently. As the model encounters new information, its rules or parameters are continuously fine-tuned. Like sketching out the frame of a building without specifying its contents, each local model transmits these parameters but not the clinical data to the global server, updating it to make the entire FL network—including other local counterparts—more optimized.

“The FL process only needs to share the model parameters optimized from each local dataset to aggregate the global model. This property allows us to keep the raw patient data unmoved from local hospitals throughout the learning process, hence effectively protecting patient privacy,” Dou explained.

Because the local models can operate independently, the FL system was easily applied to international settings. Comparable to the performance of radiologists, the model accurately detected lung injuries in chest scans from medical centers in Hong Kong, China and Germany. Moreover, diagnosis was much faster, all done and dusted in just 40 milliseconds—an evaluation process that would have taken clinicians up to 10 minutes to finish.

This high-efficiency performance also remained consistent despite differences in data formats, which varied depending on the scanning equipment and imaging protocols used.

By capturing diverse datasets, the FL technique enabled effective collaboration on a local, regional and global scale, making the model more robust and generalizable to a wide range of use cases. In this way, FL not only helped protect patient privacy, but even improved the analytical capabilities of the model.

“Privacy-preserving machine learning acts as an important enabler in such situations. It aggregates efforts from digital medical technology to provide reliable clinical assistance for timely patient care,” said Dou.





Erinne Ong reports on basic scientific discoveries and impact-oriented applications, ranging from biomedicine to artificial intelligence. She graduated with a degree in Biology from De La Salle University, Philippines.

Related Stories from Asian Scientist