January 10, 2022
In the project SyReal (Synthesizing of Realistic Data for Applicable Artificial Intelligence in Medicine), Fraunhofer Heinrich Hertz Institute (HHI) is developing methods to artificially generate realistic medical image data. To do this, the project team works with physical simulations as well as generative AI algorithms that are able to learn and realistically replicate image features. Using these artificially generated image datasets, researchers will be able to train and validate robust AI algorithms for medical applications. SyReal began its work in September 2021 and will run until August 2023. The project is funded with 1.5 million euros by the German Federal Ministry of Education and Research (BMBF).
AI-based applications can significantly reduce the burden on the healthcare system by making medical workflows easier and more efficient. To bring AI applications into healthcare, researchers need large, high-quality image datasets. Such data sets are at the heart of an AI. They are used to train the algorithms and then validate them using methods of explainable AI (XAI).
Creating these artificial image datasets is a major challenge for the use of AI in healthcare. Researchers are often denied access to medical records because they contain sensitive personal information. If the data is shared, it is often difficult to determine its origin and meta-data.
The SyReal team aims to solve this problem by developing a method for artificial image generation from medical image datasets. The researchers will initially work with image data from histopathology (fine tissue examinations, e.g. for tumor detection) and from magnetic resonance imaging (MRI) scans. The team chose these two areas because the use of AI technologies here is well researched and very promising. The applied methods can easily be extended to other medical image data types (e.g. X-ray imaging or computed tomography) later on.
To generate the images, the project team uses physical simulations as well as generative AI algorithms that can learn specific image features and then synthetically recreate them.
The "Artificial Intelligence" department at Fraunhofer HHI contributes its expertise in training, evaluating and implementing deep learning algorithms to the project. The researchers pay particular attention to the continuous auditing and evaluation of the algorithms using XAI methods developed at Fraunhofer HHI.
At the end of the project, the SyReal consortium will make the generated data available to other R&D projects free of charge to further advance the development of applicable AI.
In addition to Fraunhofer HHI, the project consortium consists of the Hasso Plattner Institute for Digital Engineering, the Max Delbrück Center for Molecular Medicine in the Helmholtz Association, the Institute of Pathology at Ludwig Maximilian University of Munich, as well as the industrial partners Aignostics, Imfusion and Dotphoton.