Modern medicine generates vast amounts of data every day—from medical records, findings, laboratory values, and physicians’ reports. This information is extremely valuable for research and the development of new AI-supported applications. At the same time, it contains highly sensitive personal data that must be carefully protected. This is precisely where the KI-AIM2 project comes in. It enables medical data to be anonymized and, in some cases, synthetically reproduced in such a way that they can be used for research and innovation without compromising patient privacy.
In the predecessor project KI-AIM, the privacy platform Cinnamon was developed for this purpose. It combines intelligent anonymization techniques with methods of synthetic data generation and automatically evaluates how private and informative the data remain. The results show that datasets prepared with Cinnamon are highly suitable for training AI-based clinical decision support systems. The software was released as open source so that academia and industry can continue to use and further develop it.
The joint project KI-AIM2 builds on this foundation and pursues three central objectives. First, Cinnamon will be connected directly to hospital and health information systems via international standards such as HL7 FHIR. This will make data anonymization easier—an important benefit for hospitals, research networks, and medical software manufacturers. Second, Cinnamon will be expanded to process not only structured data but also unstructured information such as physicians’ letters or clinical reports in a secure manner. These free-text documents account for around 80% of clinical information and contain many important details that have so far been difficult to utilize. Third, an infrastructure will be established to continuously provide anonymized data for AI systems and to transparently assess their quality.
With these enhancements, the platform used in the project will become accessible to a broad user base—from large university hospitals to smaller institutions without dedicated data protection expertise. KI-AIM2 facilitates the secure exchange of medical data, accelerates research processes, and strengthens Germany and Europe as innovation hubs in the fields of artificial intelligence and data protection. In the long term, the project will help translate medical knowledge more quickly into improved diagnostics and therapies—for the benefit of all patients.
The project is being implemented by an interdisciplinary consortium from academia, healthcare, and industry. Overall coordination is led by MeDIC in Münster. Additional project partners include: the Skin Tumor Center of the Department of Dermatology Münster; the Berlin Institute of Health at Charité; DH Healthcare GmbH (Dedalus); the Deutsches Forschungszentrum für Künstliche Intelligenz in Kaiserslautern; Health Data Technologies GmbH; TMF e.V.; and MeDIC Augsburg.
