KIRETT
The KIRETT project develops a wearable AI assistant to support emergency rescuers under extreme pressure. The device integrates and evaluates multimodal data—from dispatch, medical equipment, and paramedics—to provide real-time decision support. Using AI algorithms for situational awareness and a knowledge graph to derive action recommendations from medical guidelines, KIRETT aims to improve the quality of care, efficiency, and patient safety during critical rescue missions.
Projectdescription
The KIRETT project addressed the immense pressure on emergency paramedics by developing a wearable AI assistant designed to function as an intelligent co-pilot during critical rescue missions. The system's core was built on two research pillars: first, real-time situational awareness, which employed a multimodal machine learning based system to continuously analyze diverse data streams—from medical equipment, dispatch reports, and paramedic input—to instantly recognize life-threatening changes in a patient's condition. Second, it provided intelligent action recommendation by leveraging a sophisticated knowledge graph constructed from official medical guidelines and expert knowledge. This allowed the wearable to move beyond simple alerts and intelligently derive and suggest the most appropriate next steps, treatment alternatives, or medication dosages tailored to the specific emergency. By seamlessly integrating data, recognizing critical situations, and providing context-aware guidance, KIRETT aimed to reduce cognitive load, enhance the precision and efficiency of first aid, and ultimately improve patient survival rates. The final demonstrator was rigorously evaluated in realistic rescue exercises, which confirmed its reliability and effectiveness in the field.
Focus points/areas
-
AI in Healthcare, Embedded AI
-
Project duration
01.07.2021-30.06.2024 -
Areas
AI in Healthcare, Embedded AI -
Financing
Funding agency: Federal Ministry of Education and Research (BMBF)
Methodology
Method 1
Data preprocessing tools were first developed to process and standardize large volumes of historical rescue mission data into high-quality training datasets.
Method 2
A multimodal AI architecture was designed and trained to recognize critical patient conditions by fusing time-series sensor data with linguistic inputs.
Method 3
A comprehensive knowledge graph was constructed by applying Natural Language Processing to extract and structure action steps from official medical guidelines.
Method 4
The developed algorithms were then integrated into a wearable demonstrator and validated through practical evaluations in simulated large-scale rescue exercises.
