Imagine trying to test a new medical app, but you can't use real patient data because of strict privacy laws. This project creates 'fake' but mathematically perfect data that mimics real patients without exposing anyone's identity. It's like using a high-tech flight simulator instead of a real plane to prove a system is safe before the official inspectors arrive.
AI-driven medical devices are too complex for traditional verification, and strict privacy laws make it nearly impossible to get the large, diverse datasets needed for safe testing.
A toolkit containing scientific guidance on assurance, tools for dataset retrieval and query, and a training syllabus for regulatory compliance.
If you are a software developer dealing with the difficulty of obtaining diverse patient datasets for AI training — this project developed a toolkit that uses synthetic data to establish safety assurance before formal certification. This reduces the risk of failure during the official regulatory process.
If you are an assessor dealing with the obscurity of AI-driven device behavior — this project developed scientific guidance and tools for data-based validation. This allows for more cost-effective and high-assurance decision-making during the certification process.
If you are a data provider dealing with lengthy approval processes for anonymized data — this project developed tools for the discovery, integration, and query of multiple datasets. This enables a more dynamic and sustainable surveillance of devices throughout their lifecycle.
Based on available project data, it provides a toolkit for cost-effective decision-making and uses synthetic datasets to establish assurance before formal certification, reducing waste for regulatory bodies.
The project focuses on AI-driven devices and those using data-driven innovations, providing a toolkit for the discovery and integration of multiple datasets.
Based on available project data, the project will publish findings as public guidance and establish a public guidance work-group, though specific licensing for the tools is not detailed.
Synthetic datasets are immune to cross-referencing and harvesting of information, overcoming the lengthy approval processes required for anonymized data.
The project develops tools for the discovery, integration, and query of datasets to support through-life surveillance of medical devices.
The consortium is well-balanced for commercialization, featuring 10 partners across 7 countries. With a 40% industry ratio (including 4 SMEs), there is a strong link between the research conducted by the 3 universities and 1 research center and the actual market needs of medical device manufacturers.
Contact the Commissariat à l'énergie atomique et aux énergies alternatives in France.
Contact us to find the specific synthetic data tools developed by this consortium.
