Funding for the AIdvice digital knowledge assistant
Artificial intelligence to support cancer treatment
Around two thirds of all cancer patients look for information about their illness online. They submit questions to chatbots, scroll through forums, click their way through portals, where the information they find is often obsolete, incomplete or false. Medical specialists also make frequent use of digital resources when searching for treatments or information on correct medication dosages and face the same problems. That is the starting point for the AIdvice Assistant, an AI-supported search tool that will be developed and tested by researchers at TUM and the TUM University Hospital over the coming years. They will be supported by 1.5 million dollars in funding from the Google.org Accelerator: Generative AI. With its project, TUM was chosen as one of 20 non-profit organizations worldwide.
Individualized and understandable information
The AI-based AIdvice Assistant will combine a large language model (LLM) with a specially developed knowledge database that will contain relevant guidelines on cancer treatments. Users – whether they are patients or medical specialists – will be able to ask the chatbot about any aspect of cancer, similar to the LLMs now in common use. When answering questions, however, the chatbot will access only the evidence-based information contained in the database. This will minimize the risk of AI hallucinations, in which invented, false or misleading responses are returned.
“We don’t want the language model just to deliver answers. We want them to undergo a fact-based review. The goal is for the assistant to break down every answer into individual statements, check them against the knowledge database and link them to the original sources. Consequently, every piece of information can be traced back to the original source – for example a quote from a medical guideline,” says Florian Matthes, professor for Software Engineering for Business Information Systems at TUM.
The AIdvice Assistant is intended, on the one hand, for patients and their families looking for help with questions on such topics as treatments, side effects or prevention. But it will also relieve the workload of medical specialists who will be able to find evidence-based knowledge more quickly and access sources directly.
Use in clinical settings
“In an initial pilot phase we plan to test the chatbot in a clinical setting and steadily expand our knowledge database,” says adjunct teaching professor Dr. Jan Peeken, senior managing physician at the Department of Radiation Oncology. “The goal is then to roll out the system for Germany-wide use in clinics and family medical practices and later to other European countries.”
During the pilot phase, patients’ data will be stored on servers in the clinics where they are being treated. The data will later be stored in compliance with data protection laws on servers in Germany that meet the requirements of the healthcare sector.
- Florian Matthes is professor for Software Engineering for Business Information Systems at the TUM School for Computation, Information and Technology.
- He is also a core member of the Munich Data Science Institute.
- Dr. Jan Peeken is a senior managing physician in the field of radiation oncology and leads a research group on the topic of artificial intelligence at the TUM University Hospital.
Technical University of Munich
Corporate Communications Center
- Julia Rinner
- presse@tum.de
- Teamwebsite
Contacts to this article:
Prof. Dr. Florian Matthes
Technical University of Munich
Chair for Software Engineering for Business Information Systems
matthes@tum.de
Adjunct teaching professor Dr. Jan Peeken, PhD
TUM University Hospital
Department of Radiation Oncology
jan.peeken@tum.de
