AI speech assistance in paramedicine
Oct. 01, 2020 to Sep. 30, 2023
Task and problem definition
In Germany live approximately 21 million people with a migration background. Among them, 10 to 30 percent have low German language skills. Lack of or faulty communication in medical care can lead to undertreatment, mistreatment or even overtreatment. As a result, patients receive a lower care standard, additional costs incur and dissatisfaction among medical staff arises. Only interpreting services, available on demand and of high quality, can solve this problem.
Solution approach and technical implementation
In the project HYKIST, our researchers use artificial intelligence technologies to develop a real-time-based translation assistance system for basic and emergency medical communication. HYKIST assists interpreting language mediators through AI-assisted communication to better translate complex medical matters and technical terms during a doctor-patient conversation. For this purpose, automatic speech recognition and machine translation technologies are coupled with a dialog system for initial medical history taking and integrated into a teleconferencing platform to support the language mediators.
The HYKIST overall system combines standard telecommunication platforms with AI-based speech recognition and translation services to form an intelligent assistance system for language mediators. Speech recognition and translation is implemented through neural network structures and training methods. Almost in real time, the conversation content is mechanically recognized, translated and then presented to the language mediators as translation suggestions in the HYKIST app. The choice of the preferred translation proposal flows back as feedback into the HYKIST general system to optimize the machine translation approach. The system is supplemented by an automated initial anamnesis language assistant, which records preliminary contextual information about the patient's current situation in the emergency room in advance of the actual conversation. In HYKIST, the symbiosis of machine-based preliminary and detailed work and human input is intended to produce a learning system and to improve the quality of medical care.
Areas of application
The HYKIST overall system in progress will ultimately be evaluated in a pilot test together with clinical application partners for the area of central emergency rooms and assessed in a clinical study with regard to user acceptance. The primary application of the developed overall system will be in Germany. In addition to the target languages addressed in the project, Vietnamese and Arabic, the HYKIST platform will be expanded to include new languages such as Turkish, Polish and Russian. Due to the modular architecture of the overall system, an extension of the platform to include new use cases is very feasible. Further application areas for automated anamnesis and translation dialogs could be intensive care medicine or disaster medicine. Another possibility is also to use the developed components for translation situations in emergencies in everyday life, e.g. first aid situations.
Since September 2022, a Ukrainian version of the platform has also been in development.
• Triaphon gGmbH
• RWTH Aachen University
• AppTek GmbH