中文医疗大模型综述:进展、评估与挑战  

A Survey of Chinese Large Language Models in Medicine:Progress,Evaluation and Challenges

作　　者：籍欣萌 昝红英[1] 崔婷婷[1] 张坤丽[1] JI Xinmeng;ZAN Hongying;CUI Tingting;ZHANG Kunli(School of Computer and Artificial Intelligence,Zhengzhou University,Zhengzhou,Henan 450001,China)

机构地区：[1]郑州大学计算机与人工智能学院,河南郑州450001

出　　处：《中文信息学报》2024年第11期1-12,共12页Journal of Chinese Information Processing

基　　金：国家自然科学基金(U23A20316);河南省科技攻关项目(232102211041);郑州市协同创新重大专项(20XTZX11020)。

摘　　要：大型语言模型(Large Language Models, LLMs)已在多个领域受到广泛关注,并取得了令人瞩目的成绩。将LLMs应用于医学领域,如辅助医疗诊断、影像报告生成等,在人工智能和临床医学中都是很有前景的研究方向。由于中外医生对患者进行医疗诊断的过程存在差异,包括治疗方式、用药习惯和用药剂量等,特别是在传统中医领域,因此,为了更好地满足中文医疗领域的需求,构建大规模真实的中文医学数据集,开发中文医疗大模型是十分重要的。该文从医疗问诊、医学影像、心理健康三个方面对当前中文医疗大模型进行了概述,并介绍了当前中文医疗大模型已有的评测基准,阐述了当前中文医疗大模型面临的挑战,如幻觉、价值对齐等。未来的研究将致力于解决这些问题,并扩展医疗大模型的应用场景。Large language models have received widespread attention for their remarkable success in various fields.Applying large language models to the medical domain,such as assisting doctors in diagnosis and generating imaging reports,holds great promise for both artificial intelligence and clinical medicine.Due to differences between Chinese and foreign doctors in medical diagnosis,treatment methods,medication habits and drug dosages,it is important to build large-scale real Chinese medical datasets and develop Chinese large language models in medicine for the Chinese medical field.This paper overviews current Chinese large language models in medicine in three aspects:medical consultation,medical imaging,and mental health.Additionally,it introduces the evaluation benchmarks currently available for Chinese large language models in medicine and discusses the challenges like hallucinations and value alignment.Future researches of Chinese large language models in medicine are also discussed.

关 键 词：大语言模型 医疗大模型 人工智能 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]