AI Medical Imaging Innovation To Bypass Privacy Issues

A new AI innovation promises to overcome one of the most notable current issues with the management of medical images; the maintenance of privacy while sharing data for diagnostic purposes.

Balancing the need for the right diagnosis to be reached as quickly as possible with the data privacy rules in various jurisdictions is a challenge researchers in South Korea and the US believe they have achieved using a new AI system, Medical Xpress reports.

Researchers from Daegu Gyeongbuk Institute of Science and Technology (DGIST) worked with their counterparts from Stanford University to alleviate concerns that when data is shared with deep learning models, there is a risk of privacy violations.

The new system, called federated learning, avoids collating all data at a central server and instead only sends the learned models from each of the hospitals providing details to the server.

An additional concern is that repeated data transfers can cost time and money, but this has been addressed by building into the system limits on how often data needs to be sent. This is done not by placing a strict limit on data transfers, but by enhancing the process of model learning, which means less data needs to be sent.

Professor Sang-hyun Park of the DGIST team said: “This research will allow models to learn universally across all institutions participating in learning without sharing personal information or data.”

This will enable emerging AI technology to do what it does best – using learned knowledge to evaluate data images better and thus improve diagnostic accuracy and success.

An example of the advances this technology offers in diagnostics was recently highlighted by new research from the Institute of Cancer Research (ICT) and The Royal Marsden NHS Foundation Trust, which found that AI is nearly twice as accurate as biopsies at grading the aggressiveness of sarcomas.

This greatly improves the diagnosis and potential for successful treatment of these rare soft tissue cancers. Dr Paul Huang of the ICT said it has the “potential to transform the lives of people with sarcoma”.