Medical informatics for holistic disease models in personalized and preventive medicine
About the project
The promise of AI- and Big Data-based individualized precision medicine has recently had to be increasingly critically questioned, even though AI systems regularly match or even surpass the performance of expert medical (panels).
The quality of input data for algorithms with structural biases (“bias”), especially the inequality of medical data (e.g., the underrepresentation of population groups based on, for example, their respective gender or ethnic diversity), increases the vulnerability of AI algorithms. This leads to the shift in focus from the previous model-centric AI development to data-centric AI. This is because the quality of source data determines the goodness (“fairness”), applicability, and limitations of AI algorithms.
In the clinical environment, enormous amounts of data are already available, but they are mostly unstructured and very heterogeneous. Therefore, the focus of the MIDorAI research group is on the structured annotation and enhancement (using ontologies such as RadLex®) of radiological report texts as well as on their holistic-integrated linkage with imaging or multimodal laboratory and high-throughput data.
Another aspect of medical data that is becoming increasingly important is ensuring the privacy of this highly sensitive patient data. For this, the multimodal disease models developed by MIDorAI and the derived generation of synthetic data are expected to play a key role.
Within the junior research group, we aim to develop data integration and modeling tools as a platform and make them available within the MII and the general research community as well as Software as a Service (SaaS).
The MIDorAI research group will work on the topics of extraction of structured data from unstructured healthcare data (findings, image data), data quality and suitability of healthcare data for machine learning purposes as well as interoperable provision of disease models based on the “FAIR Guiding Principles” (Findable, Accessible, Interoperable, Reusable).
Lead of the Junior Research Group
Dr. Máté Maros
Lead of the Junior Research Group MIDorAI
Department of Biomedical Informatics | Medical Faculty Mannheim, Heidelberg University