DZHK’s flagship digitisation project Imaging "Multicentric AI development for cardiovascular disease classification and theraphy support based on imaging data"

Under current data protection and security legislation, the centralised ‘aggregation’ of medical data from multiple centres into a single central data repository is generally not permitted. The platform and infrastructure for federated learning developed in the ‘Federated Learning of TAVI Outcomes (FLOTO)’ project enables data to remain within individual clinics and algorithms to be trained in a decentralised manner. In addition to performance optimisations, the aim is to improve integration into the clinical IT landscape and to provide developers with visualisations of the algorithms to enable them to review the training process.

With the help of the existing FLOTO use case and the data already available there, these new concepts can be tested right from the start of the project and the existing platform can be optimised in terms of performance and stability.

The second sub-project focuses on developing personalised scores to classify patients with thickened heart muscle using a combination of data from cardiac magnetic resonance imaging and personal data, utilising artificial intelligence for image analysis and interpretation. On the one hand, the existing infrastructure will be used to enable the adaptation of image processing solutions to data from various centres. Secondly, the well-structured data from the UK Biobank and existing DZHK studies will be used to train the scoring models.

The project is funded by the German Centre for Cardiovascular Research and coordinated by Prof. Dr Sandy Engelhardt from Heidelberg.