The primary objective of FeatureCloud is a novel kind of medical data mining technology that – unlike existing approaches – can avoid insecure client-cloud and inter-cloud communication, and that can ensure that all data remains within (legally and technically) safe harbours: the hospitals’ IT infrastructure. We will develop, and implement a secure, novel feature communication architecture which is controlled by a tailored blockchain system for immutable access right control. Our project includes validation of this novel technology with public data, which we treat as confidential and distributed, to provide proof of principle and demonstrate that federated machine learning has the same predictive and prognostic power as traditional, centralized approaches. We will install and integrate FeatureCloud into hospital IT infrastructure and exhaustively test it using clinical trial data afterwards. FeatureCloud will run as an integrated software platform featuring an app store, which will ensure the applicability to all kinds of data types and meta data conformations – even unforeseen ones in the future.
Acceptance by patients and clinicians, but also by future app developers is the key to success. A cyclic feedback mechanism between platform and user interface design, core machine learning toolkit development, and practical testing will be employed to maximize translation into clinical practice.
FeatureCloud innovates a novel strategy to overcome the legal barrier of exchanging raw patient data and thus enable true large-scale medical data mining in a cyber-risk-minimizing manner. With FeatureCloud, we establish a platform allowing for the first time to combine globally distributed data and to learn one global computational model without the need to send any confidential data over any communication network. In addition, every patient will have full control about where and which personal data can be utilized. Having such a privacy-preserving artificial intelligence, by design, will generate rapid patient benefit as it will be able to combine data sets on a massive scale. FeatureCloud will finally allow medicine to enter the “big data” era also in practice.
In summary, FeatureCloud’s research objectives are:
- To develop a highly innovative integrated software platform and app store implementing a privacy-by-design-and-architecture approach to significantly increase cybersecurity with respect to data mining of patient data stored in hospitals and care service institutions (WP7)
- To provide proof of principle for federated machine learning fostering client-side computing as effective cybersecurity measure (WP4, WP5)
- To offer a blockchain-based data access control system taking into account cross-border requirements and usability to foster natural trust and acceptance among end-users (WP6)
- To provide proof of concept that FeatureCloud addresses the need for cybersecurity certification of products and services in the health and care domain (WP2)
- To demonstrate that FeatureCloud sets new standards for security-by-design covering the whole lifecycle of eHealth (WP3),
- To provide secure feature sharing, rather than raw data sharing, between healthcare organizations (including cross border, WP7, WP8)
- To develop a novel kind of cloud (the FeatureCloud) of significantly increased security (all WPs)
- Maximizing societal acceptance and patient trust (WP9)
On a broader scale, the FeatureCloud project will:
- Improve security of health and care data mining services by avoiding patient raw data exchange
- Significantly reduce the risk of data privacy breaches by secure federated machine learning and employing blockchain technology
- Increase patient trust and safety ensuring full control over own data, including an effective mechanism to revoke given consent
These objectives are specific, measurable, and achievable by our consortium, relevant to medicine and health IT and timed within the duration of the project to achieve several-fold proof-of-concepts in all covered technology aspects, from method development to translation into clinical practice.