Transforming Ophthalmology: How a Large Language Model Digital Patient System Improves History-Taking Skills

This study focuses on constructing an ophthalmic dataset and evaluating various large language models (LLMs) to enhance medical education through a digital patient simulation system (LLMDP). A comprehensive dataset, validated by ophthalmologists, was created from diverse sources, totaling 59,343 entries on ocular diseases. LLMs, including Baichuan-13B-Chat, were assessed for their efficiency in processing ophthalmic information. The selected LLM underwent fine-tuning using advanced techniques like LoRA for optimized performance.

The LLMDP system uses patient notes to create a knowledge base, simulating patient interactions via voice and text. It features a real-time scoring system for medical history taking, merging traditional training with advanced AI tools. Two experiments tested the system’s effectiveness and the correlation between automated and manual scoring, showcasing promising results. The randomized controlled trial, approved by an ethics committee, engaged 84 fourth-year medical students, indicating positive attitudes towards LLMs in medical education, enhancing empathy, and improving communication skills.

Source link