The Dr. von Hauner Children's Hospital (LMU Clinic) presented a coding challenge to the participants in the so-called “Makeathon” of the student initiative TUM.ai on the weekend of April 26-28 2024. Under the motto “AI in Medicine”, the aim was to come one step closer to the integration of artificial intelligence (AI) in medicine. The team led by Daniel Weiss was able to develop a proof of concept with the help of the students: It was proven that external parties can develop algorithms on non-personalized data, which can then be applied to patient data in the clinic via protected channels. Patient data is stored at the children's hospital, but still contributes to the development of artificial intelligence and thus to medical progress.
Four teams were able to successfully solve the tasks set. The Clinical Knowledge Graph developed by Prof. Matthias Mann (A knowledge graph to interpret clinical proteomics data | Nature Biotechnology) was used as the knowledge base for data processing. Using a synthetic data set, the first step was for the students to determine whether patients were healthy or ill. The disease was then classified by determining the first letter of the ICD-10 code. The codes developed were uploaded by the participating teams to the featurecloud.ai platform developed by Prof. Jan Baumbach and funded by EU Horizon 2020 (FeatureCloud – Revolutionising Cloud Communication - FeatureCloud, Privacy preserving federated machine learning and blockchaining for reduced cyber risks in a world of distributed healthcare | FeatureCloud | Project | Fact sheet | H2020 | CORDIS | European Commission (europa.eu)). These algorithms can then be used securely on the hospital server. The algorithms are trained here using real patient data, and the results are evaluated using various metrics (F1 score, accuracy). This evaluation is sent back to the developers via FeatureCloud. Constant feedback allows codes to be improved and specified. First place in the Makeathon went to the “ML4Hope” team, which achieved the highest overall metrics score. Congratulations!
In the future, this proof of concept will make it possible to develop an ecosystem for personalized medicine: while patient data is stored in a decentralized manner in compliance with data protection regulations, external parties can develop algorithms. Data can be generated and processed across Europe, constantly expanding the database. The algorithms make it possible to categorize, diagnose, compare treatment options and make decisions on a medical-individual level. There are also opportunities for the pharmaceutical industry for drug repurposing and more specific drug development. Health insurance companies can personalize preventive treatments and better calculate costs.
For us as a pediatric research center, however, the focus is on the progress of medicine. In addition to faster diagnosis and treatment options for rare diseases, data analysis can also be used to identify risk factors for diseases in adulthood.
We would especially like to thank our partners in this project. Thanks to the cooperation with the following parties, we were able to develop the concept and carry out the Makeathon:
Paul Springer and Denis Dalić from MI4People (MI4People | Machine Intelligence for Public Good | NPO)
Timm Amstein, Marius Vöhringer, Timm Hamm, Shahryar Khorasani and Jörn Kogerup from Capgemini (Transforming Health And Social Care Industry | Capgemini)
Alexander Jarasch from Neo4j (Neo4j Graph Database & Analytics | Graph Database Management System)
Julian Matschinske and Marius Klages from Dehaze (dehaze - Enabling Precision Medicine)
as well as Care-for-Rare Foundation (Care-for-Rare Foundation for children with rare diseases – Help us to help the orphans of medicine!)
Have we aroused your interest? Follow careforrare.github.io/PersonalizedMedicine/ for more information!
Or send any queries on this topic to:
Daniel.Weiss@med.uni-muenchen.de
Nicola.Goetzenberger@med.uni-muenchen.de
