Funding

1. Develop specialized image analysis AI models in a scalable manner by leveraging large language models (LLMs) to extract structured data from existing radiological reports. This will enable efficient training across various radiological tasks using unstructured clinical data. 
2. Assemble a GMAI system using a LLM as an agent that coordinates the specialized AI models, incorporates uncertainty estimates, and integrates additional tools such as clinical guidelines and laboratory values. This system will utilize multimodal inputs to support comprehensive decision-making.
3. Validate the utility of the agent-based system in realistic clinical settings by assessing its impact on diagnostic accuracy, efficiency, and the overall radiological workflow, ensuring acceptance by clinical experts.

By achieving these objectives, SAGMA will bridge the gap between current AI capabilities and the multifaceted nature of radiological practice. The project will demonstrate how an agent-based GMAI system can augment and empower human expertise, potentially transforming radiology and paving the way for similar advancements in other medical specialties.

With a consortium of leading institutions and an ambitious service portfolio, JAIF addresses the growing demand for AI solutions, bridges gaps in expertise, fosters collaboration across Europe, and strengthens Europe’s position as leader in AI-driven prosperity and growth.

As part of JAIF, JUPITER will be extended with the inference system JARVIS.

ODELIA is a unique and groundbreaking project that harnesses the power of swarm learning to revolutionize medical AI in a privacy-preserving and democratic way. With the first open source pan-European swarm learning network, we aim to develop and validate AI algorithms for breast cancer detection in MRI screening examinations, and paving the way for numerous other clinical applications.

To ensure the project’s success and deliver its transformative results, ODELIA is structured into eight distinct work packages, each focusing on specific tasks and objectives. These work packages cover everything from creating a minimum viable product to addressing regulatory frameworks and fostering communication among stakeholders. By breaking down the project into manageable components, ODELIA is poised to make a lasting impact on the medical AI landscape and improve healthcare outcomes for patients across Europe.

Liver diseases are widespread in Germany and Europe and are an increasingly important cause of sickness and mortality. Computer-aided assistance systems can use artificial intelligence (AI) to contribute to early detection and therapy decisions in liver diseases, for example by processing image and tabular data.

In this project, the partners are developing Vision Transformer (ViT, a deep learning architecture), the latest and most powerful type of artificial neural network, specifically for the diagnosis and risk prediction of liver disease. They are building on results from non-medical fields that have not yet been applied to medical research. For the first time, the researchers are enabling the systematic use of the new transformer technology for medical issues.

The consortium has a large collection of image data and associated patient data such as age, gender and comorbidities. This data will be used to train new AI models (transformers). The goal is to generate new biomarkers to predict disease progression and provide information for personalised patient treatment. These biomarkers are expected to outperform classical approaches based on conventional neural networks. In addition, TRANSFORM LIVER will provide a better understanding of disease processes in the liver, in particular cellular interactions, by applying explanatory approaches to trained transformers.

Modern cancer research is generating more extensive data sets (big data) than ever. The data originates from molecular and biochemical analyses, modern imaging procedures, clinical studies or depicts the course of a patient’s disease. These treasures of data sets need to be exploited in the future. New computer-aided approaches to use such data, namely artificial intelligence, machine learning and statistics are of great importance for the improved analysis and extraction of research-relevant information. With this funding guideline as part of the “Nationale Dekade gegen Krebs” the Federal Ministry of Education and Research (BMBF) intends to provide research groups from the field of data analysis with low-threshold access to high-quality data from translational, biomedical cancer research and routine oncological care. At the same time, researchers from the fields of data acquisition and data analysis work closely together to address clinically relevant oncological questions. In addition, the culture of data sharing for research purposes is to be promoted.

The SWAG project is developing AI methods for renal cell carcinoma that generate synthetic data from real patient data. Generative AI algorithms are trained jointly across several hospitals using swarm learning without exchanging the actual data. The resulting pseudonymized synthetic data is evaluated according to defined criteria.

This project develops and evaluates deep learning algorithms for breast cancer detection and diagnosis using MRI screening studies. By creating AI systems that support radiologists, it aims to overcome challenges in achieving acceptable diagnostic performance and transferability across clinical sites, enabling the widespread adoption of MRI-based breast cancer screening.

The aim of the project is the experimental and clinical-scientific evaluation of non-invasive assessment of cartilage functionality. The diagnosis of early osteoarthritis continues to present diagnostic difficulties despite advancing technical developments, so that scientific approaches to diagnosis in this area, not least against the background of emerging demographic developments in Western societies, are of great clinical relevance.

The overarching objective of the proposed basic research project is the development of a constitutive model that can predict from imaging data the mechanical properties of articular cartilage (i.e., cartilage functionality). If proven successful, the model-based predictions of tissue functionality based on imaging alone may be extended to other tissues and may provide a valid and substantiated framework for the non-invasive evaluation of early-stage cartilage degeneration and OA.

Metastasis is the leading cause of death in cancer patients, yet its mechanisms remain poorly understood. DECIPHER-M tackles this using multimodal foundation AI models to analyze diverse patient data, including radiological images, pathology reports, and genetic information. This approach aims to uncover how metastasis occurs, predict its likelihood and location, and identify the most effective treatments. Moreover, DECIPHER-M will develop practical tools to personalize screening and treatment for high-risk patients. These tools will help tailor therapies for individuals with metastatic disease, improving treatment effectiveness. 

CLIMB explores the potential of Contrastive Language-Image Pretraining (CLIP) for automatic breast cancer diagnosis using MRI data without manual annotation or supervision. AI has the potential to transform diagnostic imaging by accelerating scan times, improving accuracy, and reducing radiologists’ workload. However, integrating AI into clinical practice is still in its early stages and faces significant challenges. A major hurdle is the need for manually annotated data by human experts. This project aims to assess the feasibility of adapting CLIP to detect breast cancer in MRI scans by matching medical report text with visual image data. 

This project receives funding from the European Regional Development Fund (ERDF) under grant agreement No. IN-GE-3-034b