In addition to the development of techniques and tools for a comprehensive risk appraisal, SINFONIA has considered a work package (WP6) for the education and training of healthcare professionals and researchers, including an interactive and multidisciplinary Massive Open Online Course (MOOC). To facilitate the continuous education in the busy schedule of these professionals, a context-aware module (knowledge chatbot) trained on specific material for integration with the MOOC was developed by project partner Qaelum.
To this end, a Large Language Model (LLM) was used to develop a tool for fast and reliable retrieval of information. Considering the importance of accuracy and reliability of the answers, the LLM was trained in domain-specific environments, without relying on external sources such as the Internet. At first, a model with fixed technical parameters was trained and tested on the use of contrast media, showing promising results.
A Graphical User Interface (GUI) was then developed to allow users to adjust certain technical parameters, train their data models with selected educational material, test the performance by asking questions and receiving answers and, finally, score the answer to calculate the model’s accuracy. Internally at Qaelum, data models were trained and tested in different educational scenarios using relevant training material; the user manual of Qaelum’s radiation dose management system (RDMS) for medical imaging and a scientific publication on contrast and radiation dose.
The technical parameters with the best performance were adopted for the Proof-of-Concept (PoC), which was then opened to the SINFONIA partners to perform their own validation in different fields and educational scenarios using various training material. Based on the results of this assessment, the technical parameters were fine-tuned by Qaelum and the partner validation was repeated with the same material and questions. After confirming the improved performance of the fine-tuned model, the optimized technical parameters were implemented in the final version of the context-aware module, which, in its final form, was trained on material indicated by SINFONIA partners.
An Application Programming Interface (API) to integrate the contextual module with the MOOC of the SINFONIA project was also prepared.
The different stages of the task were presented at conferences. The preparation of the PoC has also been submitted to the European Radiation Protection Week (ERPW) 2024.
You may contact Niki Fitousi at niki.fitousi@qaelum.com in case of any questions on the training module.