医学大语言模型研究与应用综述

A Review on Research and Application of Medical Large Language Models

本文从模型基座、训练数据、训练方法等角度,对业界主流的医学大语言模型做了剖析;概述了医学大模型的多个能力维度,如医疗知识问答、医疗文本生成、医学文本理解、医学问诊、多模态等,并将维度与现有医学大模型的评测任务、数据集相关联;阐述了医学大模型在医学数据中心、医学决策支持、病历录入等卫生健康应用场景中可能带来的技术提升,并回顾了相关场景的研究与效果;总结了大模型当前存在的问题,对未来应用进行了展望。

This paper analyzes the mainstream medical large language models(LLMs)from the perspectives of base models,training data,and training methodologies.It summarizes several capability dimensions of medical LLMs,including medical knowledge question-answering,medical text generation,medical text comprehension,medical consultation,and multi-modal applications,and associates these dimensions with the existing evaluation tasks and datasets for medical LLMs.The paper elaborates on the potential technical advancements that LLMs could bring to healthcare settings such as medical data centers,medical decision support,and electronic health record management,and reviews the research and outcomes of LLMs’applications in these relevant healthcare scenarios.It concludes by addressing the current challenges that LLMs face and providing insights into the future.