Asan Medical Centre applies homomorphic encryption to lower breach risks

The tool was applied to a model that predicts mortality within 30 days post-surgery.

A research team from the Asan Medical Centre in South Korea has employed homomorphic encryption technology to reduce the risk of personal data breaches in artificial intelligence (AI) models.

According to the team, homomorphic encryption is a cryptographic technology that allows data analysis, computation, and modelling whilst information remains encrypted.

The tool was applied in the development of a model that predicts mortality within 30 days post-surgery.

The framework was trained using encrypted electronic medical record (EMR) data from 341,007 patients, aged 18 and older, who had undergone non-cardiac surgeries at Asan Medical Centre, Seoul National University Hospital, and Ewha Womans University Seoul Hospital.

Meanwhile, its predictive accuracy was evaluated by comparing it with a model trained solely on raw data from a single institution without using homomorphic encryption.

The encrypted, multicentre model demonstrated an accuracy rate of 95.7% in predicting postoperative mortality, compared to the raw data model's accuracy, which ranged between 88.0% and 94.2%.