Selected AI Projects

Below you will find a selection of research and development projects related to AI in medicine on which Fraunhofer MEVIS is working together with partners from medicine, science, and industry. If you are interested in further information on specific topics, please feel free to contact us.

Transforming healthcare by establishing a swarm learning network for medical AI

The goal of this EU-funded research project is to revolutionize AI in healthcare through the use of Swarm Learning (SL). It aims to overcome the obstacles of data collection in healthcare by utilising SL, where partners work together to train AI models without the need to share any personal patient data. Thereby, ODELIA will break data sharing boundaries and accelerate the scale-up of medical AI in Europe for the benefit of European citizens, patients, and clinicians.

ODELIA

Survival prediction for prostate cancer patients using federated machine learning and predictive morphological patterns

The PROSurvival project aims to more accurately predict the survival of patients with prostate cancer (PCa). The long-term goal is to generate a comprehensive, multi-site, digital dataset of PCa samples to support the collaborative development of AI for precision medicine in PCa. Previous research has shown that data from multiple sites is required to train AI models. However, these often cannot be shared due to data protection regulations. Therefore, federated AI models are being developed. Such models use patient history and clinical data in combination with publicly available data. PROSurvival will set up a privacy-compliant federated infrastructure to utilize the pool of routine clinical data. The image data will be summarized using clinically relevant pattern information, which will reduce the complexity of the dataset and facilitate analysis with commercially available hardware.

PROSurvival

Linking cancer registry data and multimodal diagnostic data for AI-based detection of biomarkers

The CanConnect project aims to link cancer registry data with a wide range of diagnostic data from the reporting clinics. The focus is on the development of a general linking concept that makes extensive, detailed case data usable for research and at the same time guarantees the protection of patient data. The feasibility and benefits of the developed linkage concept will be demonstrated using the use case of “glioblastoma”, a malignant brain tumor. For this purpose, the developed linkage concept is used to enrich cancer registry data with further diagnostic data from pathology. The linked data will be analyzed using artificial intelligence (AI) methods in order to derive new diagnostic parameters (so-called biomarkers) for glioblastoma as an exemplary application.

CanConnect

A central infrastructure for the development of data-driven AI algorithms

Modern Deep Learning methods allow AI models to be trained to automate cognitive tasks in many application areas, including digital medicine. This is a small technical revolution, because for the first time, some very demanding tasks can be meaningfully supported by computers. But even in many cases where good solutions already existed, these new AI methods again significantly improve the state of the art.

Read more...

Research meets data protection: Analyzing synthetic health data with artificial intelligence

Large quantities of high-quality data are an important basis for forward-looking research in the field of healthcare. The Health Research Data Center (Forschungsdatenzentrum,  FDZ) at the German Federal Institute for Drugs and Medical Devices (BfArM) has the task of making data from people with statutory health insurance available to authorized institutions for research purposes. There are two challenges here: On the one hand, the FDZ must be organizationally and technically capable of responding to requests in a timely and user-oriented manner. Secondly, the personal health data involved is highly sensitive and in need of protection. The “Artificial Intelligence at the Research Data Center” (KI-FDZ) project aims to make the available data accessible for research using AI methods and to allow the best possible use without being able to derive information about individual persons. Together with the Fraunhofer Institute for Digital Medicine (MEVIS), a “sandbox” system is being set up at the FDZ, i.e. a virtual space and user-friendly AI toolbox that makes it possible to test out the possibilities in a protected environment. The FDZ wants to pave the way for applications for research projects that require AI methods.

KI-FDZ

Shaping the hospital of tomorrow with artificial intelligence

Using artificial intelligence to treat patients better, reduce the workload of hospital staff and make medical processes more efficient - that is the aim of the “SmartHospital.NRW” project.

Under the leadership of University Medicine Essen, scientists from the Fraunhofer Institutes for Intelligent Analysis and Information Systems IAIS and for Digital Medicine MEVIS, RWTH Aachen University and TU Dortmund University, together with partners GSG and m.Doc, are developing concepts and solutions for how hospitals in North Rhine-Westphalia can drive forward their own digitalization and develop into “smart hospitals”. The constantly growing amount of health data makes it possible to develop intelligent and personalized applications for early health detection, diagnostics, treatment and aftercare.

SmartHospital.NRW

A central repository of digital pathology slides to boost the development of AI

To allow the fast development of AI in pathology, the BIGPICTURE project aims to create the first European, ethical and GDPR-compliant (General Data Protection Regulation), quality-controlled platform, in which both large-scale data and AI algorithms will coexist. The BIGPICTURE platform will be developed in a sustainable and inclusive way by connecting communities of pathologists, researchers, AI developers, patients, and industry parties. To achieve its goals, the BIGPICTURE project will first create the infrastructure needed to store, share and process millions of image files. Secondly, it will address legal and ethical issues to ensure patient privacy and data confidentiality issues are fully respected.

BIGPICTURE

National Research Data Infrastructure for Personal Health Data 

The goal of NFDI4Health is to merge epidemiological, public health and clinical research: a multidisciplinary team of scientists will establish a research data infrastructure for personal health data in Germany: NFDI4Health. The project is funded by the federal and state governments. The mission is to add value to research in epidemiology, public health and clinical trials. To this end, high-quality data will be made internationally accessible according to FAIR principles. NFDI4Health provides full coverage of large epidemiological studies, public health research, and investigator-initiated clinical trials in Germany, as well as co-development of NFDI4Health with the user community. As part of the COVID-19 Task Force and the Radiomics / Imaging AI use case, Fraunhofer MEVIS is working on AI-based image analysis and federated learning.

NFDI4Health

Digitized data acquisition and AI for safe operations

KIPeriOP is a research project funded by the German Federal Ministry of Health (BMG) with the aim of improving perioperative risk management and reducing perioperative mortality and permanent damage. Clinical guidelines already support perioperative decision-making and will be complemented in the project by the trustworthy use of artificial intelligence, including the prediction of postoperative risks based on preoperative risk factors. The project consortium combines outstanding clinical, technical, ethical and economic expertise and is led by the University Hospital of Würzburg (clinical coordination) and the Fraunhofer Institute for Digital Medicine MEVIS (technical coordination).

KIPeriOP

The German-wide RAdiological COOperative Network

The COVID-19 Pandemic Radiological Cooperative Network RACOON is a joint project of the radiology departments at all 36 German university hospitals. Experts at all sites segment and annotate a large number of lung CT scans in a structured and uniform manner. The resulting data is used to develop medical assistance systems and an early warning system based on AI. Fraunhofer MEVIS is one of the three technical partners in RACOON, in addition to the German Cancer Research Center (DKFZ) and the Technical University of Darmstadt.

RACOON applies MEVIS technology called CuraMate (formerly SATORI) for interactive data curation, AI, and radiomics analysis.

RACOON

EcosysteM for Pathology Diagnostics with AI Assistance

Recent advancements in image-based diagnostics, driven by artificial intelligence (AI) methods, have been significant. EMPAIA International e. V. is committed to facilitating routine use of validated and certified AI solutions by healthcare professionals. Additionally, the promotion of AI usage involves the proactive elimination of regulatory, legal, technical, and organizational obstacles.

The inadequate standardisation of interfaces in digital pathology is an important obstacle on the way to more digitalisation. Building on established standards such as HL7 and DICOM, we develop and disseminate open standards for the rapid dissemination of modern methods in clinical diagnostics. This also serves to improve patient access to their data.

EMPAIA

Computer-aided automatic prediction of the development of malignant cerebral edema after medial infarction

The aim of the APICES project is to use the “machine learning” method to analyze computer tomographic images (CT images) and clinical data from 1,500 patients and to develop a model that helps to detect brain swelling at an early stage and predict its progression. First, computer-based algorithms automatically identify characteristic features from the CT images and clinical data (learning phase). In a subsequent validation phase, the model developed in this way is checked against new data sets. The aim of using machine learning is to better understand brain swelling and detect it at an early stage.

APICES

Deep learning for translational research in glioma surgery and stroke imaging

The “Translational Brain Imaging Training Network” (TRABIT) is an interdisciplinary and intersectoral joint effort of computational scientists, clinicians, and the industry in the field of neuroimaging. Its aim is to train a new generation of innovative and entrepreneurial researchers to bring quantitative image computing methods into the clinic, enabling improved healthcare delivery to patients with brain disease.

Since brain imaging often visualizes disease effects with much greater sensitivity than clinical observation, it holds great promise to help diagnose patients at the earliest stages of their disease, when treatment is most effective; and personalize their treatment by evaluating their response to a specific intervention. A fundamental bottleneck in translating the wealth of information contained in medical images into optimized patient care is the lack of patient-specific computational tools to help analyze and quantify the torrent of acquired imaging data. The last two decades of medical image computing research have matured to allow robust and automatic assessment of carefully homogenized scientific studies of mostly healthy brain scans. Yet analyzing the “wild” type of neuroimaging data arising in the standard clinical treatment of brain disorders remains a hard and unsolved problem.

TRABIT