arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2512.01210cs.AI
收藏

基于知识图谱的大型语言模型用于疾病预测

Knowledge Graph Augmented Large Language Models for Disease Prediction

Department of Computer Science, Emory University(埃默里大学计算机科学系);Division of Medical Sciences, Oxford University(牛津大学医学系);Department of Computer Science and Engineering, UCSC(UCSC计算机科学与工程系);Nell Hodgson Woodruff School of Nursing, Emory University(埃默里大学内尔·霍根·伍德鲁夫护理学院);School of Medicine, Emory University(埃默里大学医学院);Rollins School of Public Health, Emory University(埃默里大学罗林斯公共卫生学院)

Ruiyu Wang, Tuan Vinh, Ran Xu, Yuyin Zhou, Jiaying Lu, Francisco Pasquel, Mohammed K Ali, Carl Yang

AI总结:

本文提出基于知识图谱的大型语言模型框架,用于提升疾病预测的准确性与解释性,通过构建时间一致的推理依据,在多个数据集上取得优于经典基线的性能。

摘要:

电子健康记录(EHRs)能够实现强大的临床预测,但解释通常较为粗略,难以用于患者层面的决策。我们提出了一种知识图谱(KG)引导的思考链(CoT)框架,用于在MIMIC-III数据集上进行就诊层面的疾病预测。我们将ICD-9代码映射到PrimeKG,挖掘与疾病相关的节点和路径,并利用这些路径构建时间一致的CoT推理依据,仅保留结论与观察结果一致的样本。我们对两种小型队列(400和1000个索引就诊)中的十种PrimeKG映射疾病进行了轻量级指令微调的LLM(LLaMA-3.1-Instruct-8B和Gemma-7B)。我们的模型优于强大的经典基线,达到AUROC 0.66-0.70和macro-AUPR 0.40-0.47。在没有额外训练的情况下,模型能够零样本迁移至CRADLE队列,将准确性从0.40-0.51提高到0.72-0.77。在盲测临床研究中,KG引导的CoT推理依据在清晰性、相关性和正确性方面始终受到偏好。代码可在:https://github.com/JonathanWry/KG-guided-LLM-pipeline获取。

英文摘要:

Electronic health records (EHRs) enable strong clinical prediction, but explanations are often coarse and hard to use for patient-level decisions. We propose a knowledge graph (KG)-guided chain-of-thought (CoT) framework for visit-level disease prediction on MIMIC-III. We map ICD-9 codes to PrimeKG, mine disease-relevant nodes and paths, and use these paths to scaffold temporally consistent CoT rationales, retaining only samples whose conclusions match observed outcomes. We fine-tune lightweight instruction-tuned LLMs (LLaMA-3.1-Instruct-8B and Gemma-7B) on two small cohorts (400 and 1,000 index visits) across ten PrimeKG-mapped diseases. Our models outperform strong classical baselines, reaching AUROC 0.66-0.70 and macro-AUPR 0.40-0.47. Without additional training, the models transfer zero-shot to the CRADLE cohort, improving accuracy from 0.40-0.51 to 0.72-0.77. In a blinded clinician study, KG-guided CoT rationales are consistently preferred for clarity, relevance, and correctness. Code is available at: https://github.com/JonathanWry/KG-guided-LLM-pipeline

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑