用指令调整的大型语言模型来提取临床数据:从CT放射学报告创建大动脉测量数据库
Ely Erez1, Sedem Dankwa1, McKenzie Tuttle1
1Division of Cardiac Surgery, Yale School of Medicine, New Haven, CT USA.
Journal of healthcare informatics research
|November 13, 2025
概括
指示调整的大型语言模型有效地从胸部CT报告中提取大动脉直径. 这种高准确度的数据提取通过使测量可访问,有助于临床决策和研究.
科学领域:
- 医疗信息学 医疗信息学
- 人工智能在医学中的应用
- 放射学 信息学 信息学
背景情况:
- 胸部计算机断层扫描 (CT) 报告包含重要的大动脉测量,但它们通常是无结构的.
- 这些测量的可访问性有限,阻碍了临床决策和对大动脉疾病的研究.
- 结构化数据提取可以改善患者护理,并促进人口研究.
研究的目的:
- 为了比较大型语言模型 (LLM) 与从胸部CT报告中提取大动脉直径的传统模型的性能.
- 评估指令调整,少数拍摄和零拍摄的拉玛模型与微调的BERT模型相比.
- 评估使用LLM用于高保真度医疗信息提取的可行性.
主要方法:
- 使用了356690份胸部CT报告 (2013-2023) 的数据集.
- 2010年报告的一个子集在八个解剖部位手动注释了大动脉直径.
- 调整指令,少量射击和零射击的Llama 3.1模型与微调的临床BERT模型进行了比较.
主要成果:
- 调整为指令的Llama 3.1模型获得了最高的性能,测试组F1得分为0.970.
- 这种模型的性能优于少数射击 (F1=0.838),零射击 (F1=0.663) 拉玛模型和临床BERT (F1=0.954).
- 该指令调整模型在整个数据集的13.85%中确定了大动脉测量.
结论:
- 调整指令的LLM能够从无结构的放射学报告中高准确地提取大动脉测量,并提供最小的注释.
- 这种方法显著提高了临床决策和研究关键数据的可访问性.
- 开发的框架可以适应各种医疗信息提取任务.
相关概念视频
Imaging Studies for Cardiovascular System V: CT
268
Cardiac computed tomography (CT) scanning is an advanced cardiac imaging technique that utilizes CT technology, with or without intravenous (IV) contrast, to produce accurate cross-sectional virtual slices of specific areas of the heart, coronary circulation, and major blood vessels such as the aorta, pulmonary veins, and arteries. The computer processes these slices to generate three-dimensional images. Multidetector CT (MDCT) is a rapid form of CT scanning that captures multiple slices...
268
Imaging Studies for Cardiovascular System VI: Calcium -Scoring CT
377
Calcium-Scoring CT ScanA calcium-scoring CT scan, also known as coronary artery calcium (CAC) scan, detects calcium deposits in the coronary arteries. This test assesses the risk of coronary artery disease (CAD), which can lead to cardiovascular events such as angina, heart failure, and sudden cardiac arrest.A calcium-scoring CT scan is generally recommended for individuals at intermediate risk of CAD without symptoms. It includes:Men aged 40-75 and women aged 50-75: Especially those with a...
377


