An AI-assisted, failure mode-based toolkit for proactive risk management in radiotherapy: A feasibility study

Background and purpose: Proactive risk management is essential for safe radiotherapy (RT), but formal approaches such as Failure Modes and Effects Analysis (FMEA) can be difficult to implement routinely because of time and resource constraints. We developed i-SART (Intelligent Safety Assistant for Radiotherapy), an AI-assisted, failure mode-based toolkit to support proactive risk management and safety activities in RT, and conducted a feasibility evaluation. Materials and methods: i-SART is a web-based prototype combining a harmonised failure mode (FM) database derived from 10 FMEA studies, an online portal for FM submissions and a conversational assistant (chatbot) configured to provide structured FM analyses, mitigation suggestions and illustrative incident examples. Internal testing assessed response structure, accuracy and hallucinations. An online survey of RT professionals explored usability and perceived value. Results: The database contained 419 FMs spanning the RT workflow. The assistant produced structured, clinically plausible mitigation suggestions but was only partially accurate, with hallucinations mainly affecting guideline references and incident details. Fifty-one professionals from 11 countries completed a survey, giving median ratings of 4/5 across domains, including ease of use, support for proactive risk management, and perceived error-reduction potential. Conclusion: This feasibility study shows that a failure mode-based toolkit combining a curated database, submission portal, and conversational assistant is technically achievable and was rated favourably by early users for supporting risk management activities. Further work should expand the database, strengthen safeguards against hallucinations, and evaluate effects on risk assessment processes and patient safety.