2578 - RadOnc Smart Learn: An AI-Powered Platform for Longitudinal, Personalized Oral Board Preparation in Radiation Oncology
Presenter(s)
A. M. Conteh1, C. Parikh2, R. T. Hoppe3, and M. K. Buyyounouski3; 1Department of Radiation Oncology, Stanford University, Stanford, CA, 2Thomas Jefferson University, Philadelphia, PA, United States, 3Department of Radiation Oncology, Stanford University School of Medicine, Stanford, CA
Purpose/Objective(s): Oral board examinations in radiation oncology demand comprehensive clinical knowledge, real-time decision-making, and effective communication under high-pressure conditions. Traditional exam-preparation resources— mock orals, static question banks, and geographically constrained workshops—often fail to provide consistent, adaptable practice throughout a resident’s training. We hypothesize that an AI-driven platform offering realistic interactive cases and evidence-based feedback is feasible, scalable and widely accessible from early clinical exposure through final preparation.
Materials/Methods: We developed a web-based platform using a Django backend for data management, user authentication, and content delivery. The frontend employs JavaScript and HTML with Bootstrap for a responsive interface. Multiple disease site–specific AI agents—each aligned with a major radiation oncology domain (e.g., breast, prostate, head and neck, lung)—are powered by an advanced large language model (Gemini 1.5) on Google Vertex AI. Real-time retrieval of established guidelines is achieved via the Vertex AI retrieval-augmented generation (RAG) Engine, which incorporates standard practice frameworks, including ASTRO guidelines, NCCN protocols, and institutional references, to ensure that case content is evidence-based. To enable interactive oral board simulations, the platform integrates speech-to-text (Deepgram NOVA2-med) and text-to-speech (Azure Cognitive Services). This functionality allows users to verbally engage in simulated exam scenarios and receive immediate audio feedback. The system is hosted on Azure App Service for scalability and secure deployment.
Results: A fully functional prototype, RadOnc Smart Learn, provides two main modes of engagement. Tutoring Mode allows guided learning with detailed explanations and clarifications, while Mock Oral Mode simulates board-style examinations with timed sessions and AI-generated cases. Scenarios adapt to evidence-based guidelines, and users receive personalized, on-the-spot feedback aimed at improving clinical reasoning and communication.
Conclusion: RadOnc Smart Learn addresses the need for an accessible, adaptable, and evidence-based approach to oral board exam preparation in radiation oncology. By integrating evidence-based case generation, adaptive learning pathways, and immediate feedback, this platform has the potential to enhance not only exam performance but also clinical competence among radiation oncology trainees. A planned pilot study with radiation oncology residents will assess the RSL’s impact on knowledge acquisition, retention, and overall exam readiness through pre/post testing and comparative analyses. To experience the prototype platform’s interactive features firsthand, visit https://radoncsmartlearn.com/