First Artificial Intelligence System to Support Clinical Decision Making for Patients with Vestibular Schwannomas

We built an artificial intelligence tool to support patient knowledge and support the doctor-patient interaction for decision-making in vestibular schwannoma care. Our goal was to build a system based on available data between care the alternatives of observation, resection and radiosurgery, using established high-level literature comparing options and outcomes. Such a system would provide individual patient care recommendations and allow questions and interaction with the knowledge foundation. The 4 elements of the system include (1) a curated literature focused on alternatives and outcomes with modern comparison data; (2) individual patient and tumor specific characteristics; (3) a priority listing for desired patient outcomes - tumor response, facial function, hearing, balance and tinnitus, and (4) the AI generated care recommendation with reasoning and references. The system was built on data from 14 articles entered into our language model. Age, sex, MRI date(s), maximum tumor size(s), Koos grade, Gardner Robertson hearing classification (using speech discrimination score or pure-tone average, whichever was better), facial weakness (House-Brackmann grading), balance status (normal, mild, moderate, or severe symptoms), and ranked patient priorities (durable tumor control, complete tumor removal, preserve facial movement, preserve balance, preserve hearing, tinnitus improvement) were input into the literature-trained AI model to generate a treatment recommendation. We first tested the utility of the recommendations by reviewing actual care in 77 radiosurgery and resection patients. The concordance between AI and neurosurgeon treatment decisions was 82%. We also compared the AI reasoned report to the actual patient’s clinical chart documentation. The surgeon or patient can interact with the system and ask other questions as desired. Conclusion: Our AI system guides decision making based on a foundation of high-quality medical literature coupled with specific patient and tumor characteristics and the patients own goals. The system is modular and easily modified by adding or subtracting individual articles and modifying the patients own data elements. It can be used to justify care plans for individual patients, to enhance the patients understanding of their options, and to elevate the doctor-patient interactive experience. Publication History Article published online: 27 February 2026 © 2026. Thieme. All rights reserved. Georg Thieme Verlag KG Oswald-Hesse-Straße 50, 70469 Stuttgart, Germany