ESTRO 2025 - Abstract Book

S2178

Interdisciplinary – Education in radiation oncology

ESTRO 2025

Results: From December 2022 to October 2024, we organized 10 scientific meetings covering a wide range of topics both in terms of techniques and tumor locations. We addressed subjects such as “Medical applications of proton therapy in head and neck cancers management”, “Prescription methods and calculation algorithms in thoracic stereotactic radiotherapy”, “State of the art FLASH therapy for dermatological tumors”, radiobiology related issues, etc. Therefore, these meetings have enabled us to update our internal treatment guidelines, particularly for thoracic tumors (1-fraction stereotactic treatment, choice of prescription isodose, management of ultracentral tumors). Additionally, this collaborative and multidisciplinary approach has fostered a "speak-up" culture within the department, where open dialogue and constructive feedback are encouraged across disciplines and hierarchical levels. Conclusion: This horizontal training model, led by junior physicians and centered on the physician-physicist pair, has thus proven effective in enhancing the dynamic of continuous learning and innovation integration in oncology radiotherapy, while also fostering interdisciplinary cooperation, particularly with service physicists, and improving clinical practices.

Keywords: Physician – physicist pair , scientific meeting

References: 1. Glassick CE (2000, Sep). Boyer’s expanded definitions of scholarship, the standards for assessing scholarship, and the elusiveness of the scholarship of teaching. Academic Medicine, 75(9), 877–880. 10.1097/00001888-200009000 00007

2129

Digital Poster MedPhys Chat: A retrieval augmented generation (RAG) pipeline focusing on AAPM reports using the latest large language models. Hossein Jafarzadeh 1 , Jonathan Kalinowski 1 , Farhood Farahnak 1 , Shirin A. Enger 1,2 1 Medical Physics Unit, McGill University, Montreal, Canada. 2 Lady Davis Institute, Jewish General Hospital, Montreal, Canada Purpose/Objective: The reports from the American Association of Physicists in Medicine (AAPM) serve as important guidelines in daily clinical practice and research in the field of medical physics. We prepared a ChatBot that uses retrieval augmented generation (RAG) and the latest large language models (LLM) to answer questions regarding AAPM reports. A total of 194 AAPM reports were downloaded in PDF format. The text was converted to XML and divided into 1000 character segments with a 200-character overlap and converted into numerical vectors using embedding generators from OpenAI, LLaMA, and Vertex AI. These vectors were then stored in a Chroma vector database. Running Inference When a user query is received, an embedding is generated, and the cosine similarity metric identifies the 5 closest AAPM embeddings from the database. The query and corresponding texts for these embeddings are presented to each LLM. The models used were GPT-4O-mini, LLaMA-3.1-70B, LLaMA-3.2-3B, and Google Gemini-1.5-Flash, each tested with default temperature settings and a temperature of 0.1. Evaluation Material/Methods: Data Preprocessing