T5.6 Use Case “Radiomics / Imaging AI”

The Use case “Radiomics / Imaging AI” is led by the Fraunhofer MEVIS. The BIPS, the Fraunhofer FIT, the University of Cologne and the Leipzig University are involved as project partners.

Biomedical imaging has become an important area of data collection in epidemiologic and clinical studies because of its ability to detect both local and systemic changes, its high information density and its wide availability. Radiomics is defined as a method that uses pattern recognition and machine learning algorithms to extract large amounts of quantitative features from medical images.

Modern machine learning techniques, especially deep learning, combined with conventional data analysis within multidisciplinary healthcare data, promise to bring medical imaging closer to understanding complex diseases in a quantitative way by integrating clinical, structural, and imaging-based features.

Radiomics discovers relationships between image features and clinical data. Copyright: Fraunhofer MEVIS

In collaboration with TA3 “Services”, NFDI4Health infrastructure components will be implemented in this Use Case as a prototypical AI analysis test bed along the Radiomics paradigm, with the ultimate goal of standardising and validating AI for wider use. Where appropriate, the DICOM standard will be used to harmonise image data and associated metadata in accordance with TA2 “Standards for FAIR Data”.

In particular, a proof-of-concept for automated AI-supported quality assurance of DICOM image data will be implemented and evaluated in sub-cohorts of heterogeneous MR volume data as part of NFDI4Health. In addition, a demonstrator will be created that enables predictive pattern recognition in combined imaging and structured epidemiological data.

The long-term goal within NFDI4Health is to provide an AI-based pilot radiomics platform for the automated quality assurance and analysis of biomedical (image) data as a service for the community. The use of AI allows for easy adaptability to new application contexts and continuous optimisation, particularly through federated learning.