聊天GPT-4o和四个开源大型语言模型在基于中国罕见疾病目录的诊断生成中的性能:比较研究
Wei Zhong1, YiFan Liu1, Yan Liu1
1Department of Prenatal Diagnosis, Beijing Obstetrics and Gynecology Hospital, Capital Medical University, Beijing Maternal and Child Health Care Hospital, No. 251 Yaojiayuan Road, Chaoyang District, Beijing, China, 8618810963279.
Journal of medical Internet research
|June 18, 2025
概括
聊天GPT-4o实现了罕见疾病的最高诊断准确度. 整合检索增强生成 (RAG) 显著改善了开源大语言模型 (LLM),突出了基于语言和参数化的临床使用需要仔细选择模型的需要.
科学领域:
- 人工智能在医学中的应用
- 临床决策支持系统 临床决策支持系统
- 罕见疾病的诊断 罕见疾病的诊断
背景情况:
- 诊断罕见疾病是复杂的,通常受到医生知识的限制.
- 大型语言模型 (LLM) 提供了一种新的方法来增强诊断工作流程.
研究的目的:
- 评估ChatGPT-4o和罕见疾病的开源LLM的诊断准确性.
- 评估语言 (英语与中文) 对LLM诊断绩效的影响.
- 探索检索增强生成 (RAG) 和思维链 (CoT) 推理的有效性.
主要方法:
- 从中国罕见病目录中提取了121种罕见疾病的临床表现.
- 评估了ChatGPT-4o和四个开源的LLM (qwen2.5:7b,Llama3.1:8b,qwen2.5:72b,Llama3.1:70b) 在英语和中文.
- 使用RAG和Cot重新评估了表现最差的模型;使用麦克纳马测试比较准确度;对罕见疾病熟悉度的临床医生进行了调查.
主要成果:
- 聊天GPT-4o实现了最高的诊断准确率 (90.1%).
- 语言显著影响了LLM的表现;Llama3.1:8b在英语中表现更好,而qwen2.5:7b在两种语言中表现相似.
- RAG将qwen2.5:7b精度提高到79.3%,检索精度达到85.9%;临床医生调查显示,罕见疾病管理方面的知识缺口.
结论:
- 聊天GPT-4o在罕见疾病诊断方面表现出卓越的诊断性能.
- 开源的LLM需要更大的参数才能在中文中具有可比的准确性;RAG集成增强了诊断实用性,特别是在资源有限的环境中.
- 优先考虑临床LLM实施的语言相关性和精心策划的知识库,谨慎使用低参数模型.
相关概念视频
Classification of Illness
8.0K
The meaning of illness is individualized to each person who experiences an alteration in health. In contrast, disease is a medical term indicating a pathological change in the structure and function of the body or mind. It is a condition that has specific symptoms and boundaries.
An illness is a response to a disease in which the person's level of functioning is changed compared with a previous level. The general classification of illness includes acute and chronic.
Acute illness is severe...
An illness is a response to a disease in which the person's level of functioning is changed compared with a previous level. The general classification of illness includes acute and chronic.
Acute illness is severe...
8.0K
Single Nucleotide Polymorphisms-SNPs
16.0K
A single nucleotide polymorphism or SNP is a single nucleotide variation at a specific genomic position in a large population. It is the most prevalent type of sequence variation found in the human genome. Point mutations that occur in more than 1% of the population qualify as SNPs. These are present once every 1000 nucleotides on an average in the human genome. Replacement of a purine with another purine (A/G) or a pyrimidine with another pyrimidine (C/T) is known as a transition. In contrast,...
16.0K


