Innovative Learning: Using Artificial Intelligence to Navigate Simulated Neuro-Ophthalmology Cases - 5643
Mon statut pour la session
Author’s Disclosure Block: Samir Touma: none; Paul Touma: none; Moncef Amchech: none; Tracy Aoun: none; Lahoud Touma: none; Katie Luneau: none
Abstract Body
Purpose: Neuro-ophthalmology is one of the most complex and enigmatic branches of ophthalmology. Due to the rarity of certain conditions, it can be challenging for residents to encounter patients with these pathologies. Furthermore, obtaining a thorough medical history is crucial, and asking the right questions is essential for making accurate diagnoses. This study aims to evaluate the effectiveness of an online platform that leverages large language models (LLM) to simulate clinical cases for ophthalmology and neurology residents. Study Design: Proof-of-concept study Methods: An online platform was developed by two software engineering students, utilizing the GPT-4 model (OpenAI) with customized instructions to simulate virtual patients. The instructions sent to the model included key elements such as past medical history, family history, social history, symptoms, medications, and allergies.Simulated cases coveringneuro-ophthalmological pathologies were developed.Users interact with these virtual patients through a chat interface, asking clinical questions such as medical history and current symptoms. They then complete a virtual physical examination and perform additional tests (e.g., OCT, imaging, blood test) as needed, using the platform’s interactive features. Finally, userssubmit their diagnoses, which will be either confirmed or corrected by the AI. Upon submitting the correct diagnosis, users will receive a brief teaching section. Results: Ten simulated cases covering rare neuro-ophthalmological conditions were produced,includinggiant cell arteritis, acquired achromatopsia and optic neuritis. Each case takes between 5 to 10 minutes to complete. Users found the interface intuitive and appreciated the feedback and teaching provided after submitting their diagnosis. Virtual patients accurately simulated rare conditions, allowing residents to ask relevant clinical questions. The system successfully processed the clinical questions and provided appropriate responses. Conclusion: LLM--driven virtual patient simulations represent a promising, innovative approach to supplement traditional teaching methods particularly for rare and complex neuro-ophthalmological conditions. Further data from resident feedback collected through a survery will provide additional insights into the educational value of this platform.