SyReal

Many steps in everyday healthcare are labor-intensive and time-consuming. Due to the increasing burden on the healthcare system, artificial intelligence (AI)-based algorithms are currently being developed in order to make medical workflows more efficient. This project develops solutions to the following major challenges:

1. Medical data is often not available to a sufficient extent for the development of algorithms due to data protection reasons.
2. AI algorithms in medicine need to be particularly reliable, so that they can also recognize rare disease characteristics as well as being robust to external interfering factors so as not to cause harm to humans.

Currently, the reliability of AI algorithms in medicine is often impossible to prove and therefore often prevents the use of AI applications in practice. In most cases, there is a lack of a comprehensive data base that realistically represents different test cases. The goal of SyReal is to generate realistic data for applicable AI in medicine.

To generate the medical image data, this project will use and further develop deep-learning based generative AI algorithms (GANs), which learn and realistically replicate image features. In our project, we bring together different academic institutions and partners from the commercial sector with interdisciplinary competencies from computer science, AI, statistics, physics, and medicine to jointly enable the generation of realistic image data in medicine. Initially, the project focuses on the generation of realistic histological and magnetic resonance image data, which will be extended to other types of medical image data (e.g. computed tomography).

Our project plan has three overall goals:

1. to develop methods for the artificial generation of medical images that realistically represent everyday clinical practice.
2. to make available, free of charge, synthesized MR and histology image data that can be used by other researchers and developers to create robust AI algorithms.
3. integration of the developed methods for image data generation into a software with permissive license and marketing by our SME project partners.

• Hasso-Plattner-Institute
• Fraunhofer HHI
• Max Delbrück Center for Molecular Medicine
• Pathologisches Institut der Ludwig-Maximilians-Universität München
• Aignostics
• ImFusion
• Dotphoton