Ontology-driven Large Language Model for Surgery (OntoSPM-LLM)

Focus:	Talents
Type of funding:	Individual funding programmes
Programme:	CZS research boost
Funded institution:	Stuttgart University of Applied Sciences

Prof. Darko Katic, Professor of Robotics and Artificial Intelligence at HfT Stuttgart, is developing an AI-supported assistance system for surgery. The system uses large language models and retrieval augmented generation and uses medical ontologies to make surgical procedures understandable.

Goals

Surgical training and practice have traditionally been based on experience, textbooks and case studies. Current AI-supported systems still have considerable weaknesses for use in the medical field. Lack of reliability and explainability lead to inaccurate or incorrect conclusions. AI models hardly combine medical ontologies. Therefore, there is a need for an AI system that generates medically validated answers.

The "OntoSPM-LLM" project is developing a system that combines Large Language Models (LLMs) and Retrieval Augmented Generation (RAG), access to external knowledge sources, to create a semantically sound knowledge base for surgery.

In the long term, this knowledge could also be used for physical assistance in the operating room using robotics. The use of ontologies ensures that the AI answers are consistent, explainable and evidence-based. The project is a first step in this direction and is developing a natural language chat system for medicine. This will enable surgeons to discuss medical issues with a competent AI, deepen their knowledge and, ideally, make better decisions.