A recent study conducted by researchers at Asan Medical Center in South Korea investigated the use of homomorphic encryption (HE) to protect patient data used in training AI models.
HE is a cryptographic scheme that allows computations to be performed on encrypted data, which is particularly useful in scenarios requiring data privacy, such as secure elections and, in this case, healthcare.
The researchers aimed to assess the viability of HE in preserving privacy and security when combining large datasets from multiple institutions for AI model training.
The study, published in *JMIR Medical Informatics*, involved the use of electronic medical record (EMR) data from over 300,000 patients across three major hospitals: Asan Medical Center, Seoul National University Hospital, and Ewha Women’s University Seoul Hospital.
The data was encrypted using HE and then used to train an AI model designed to predict mortality rates within 30 days following surgery. The findings demonstrated that HE could be effectively applied to aggregate data from various sources while maintaining strict privacy and security protocols.
One significant outcome of the study is the potential for smaller hospitals to utilize HE to develop their own AI models by safely accessing data from larger institutions. This approach could enhance the predictive accuracy of AI models beyond what could be achieved with data from a single institution alone.
The ability to combine encrypted data securely allows for a broader and more diverse dataset, which is crucial for improving the reliability of AI predictions.
The relevance of this study extends beyond healthcare, as HE has been explored in various sectors, including finance and IT, to enhance data privacy. In healthcare, the challenge of obtaining large and diverse datasets for AI model training is often complicated by strict privacy regulations.
The research highlights HE as a solution to these challenges, offering strong security for personal information and enabling predictive modeling on encrypted data, thus addressing privacy concerns.
The broader trend in South Korea includes government support for AI research in the medical field. The Ministry of Health and Welfare has launched initiatives such as the Medical Data Utilisation Project, which facilitates collaboration between researchers and hospitals, allowing for secure access to medical datasets.
This project is part of a larger effort to advance AI in healthcare, with several hospitals designated as centers for safe medical data utilization.
