Truly secure ways of storing and processing primary medical data are in high demand

One of the most pressing problems in training powerful artificial intelligence (AI) and machine learning in health informatics is that primary medical data is usually only stored locally at each hospital, but that it needs to be made accessible in anonymized form (for health care professionals and scientific research) in one central repository, i.e. a central cloud. Any centralized data storage, however, also represents a “central point of attack”. In light of enormous data leaks in the recent past, sensitized public opinion, and dwindling patient trust, truly secure ways of storing and processing primary medical information are therefore in high demand.

The European Network and Information Security Directive (NISD) entered into force in August 2016, and the European General Data Protection Regulations (GDPR) took effect on May 25th 2018. Since then, the entire legal field, European data providers, scientists, and information technology (IT) specialists are challenged to find ways of giving patients and hospitals full control over how their sensitive patient data is processed, including mechanisms to effectively revoke a previously given patient’s consent.

The FeatureCloud research consortium will rise to this challenge by developing the worldwide first “privacy by design” solution. We are about to create an AI platform – the “FeatureCloud” – based on a ground-breaking, novel cloud infrastructure that does not require transfer of any sensitive raw data to a centralized cloud. Instead, it only exchanges the learned representations which are totally anonymous by default. Local federated machine learning at each hospital, blockchain technology, and application of strict privacy ethics will ensure that patients maintain full and immutable control over where and which of their data can be accessed.   

placeholder