文献速递腹部CT

大语言模型基于CT自由文本报告对肝脏病变分类的多中心真实世界分析

Characterising liver lesions from free-text computer tomography reports - A real-world multicentre analysis.
PubMed 文献 · Eur J Radiol · 2026-01-17

本研究利用5家机构的296份腹部CT自由文本报告,评估通用大语言模型GPT-4与医学微调大语言模型Med-LM对肝脏病变LI-RADS分类的性能。结果显示,Med-LM使用优化提示时表现最佳,但病变层面LI-RADS完全分类准确率仅61.3%,良恶性二分类准确率升至82.3%。高误分类率与不一致的可重复性限制了其临床应用。

要点

  • 大语言模型从无结构CT报告中识别肝脏恶性病变的能力优于进行完整的LI-RADS分类。
  • 医学微调的大语言模型结合优化提示的病变分类性能显著优于通用模型,但仍存在高误分类率。
  • 良性病变和LI-RADS低级别分类的非分类率与误分类风险较高,不宜直接据此做临床决策。

核心结论

  • Med-LM在基于无结构CT报告分类肝脏病变方面优于GPT-4,但两者均存在高误分类率和不可接受的非一致性,目前不宜用于临床。

研究设计与样本

  • 多中心回顾性研究,纳入2014-2020年来自5家机构、共包含654个病灶的296份腹部CT自由文本报告。
  • 以两名放射科医生协商一致的LI-RADS评分为参考标准,采用简单提示与优化提示两种策略进行病变及患者水平分析。
  • 同时使用5分李克特量表评估报告质量,并计算了模型的非分类率、准确率和重复性。

诊断标准与关键数值

  • 病变层面LI-RADS完全分类准确率最高为Med-LM优化提示的61.3%,患者层面最高为52.4%。
  • 病变层面良恶性二分类准确率最高为Med-LM优化提示的82.3%,患者层面最高为86.5%。
  • Med-LM优化提示在所有分析中表现最优,并显著优于其他模型(所有 P < 0.001)。
  • 非分类率介于12.7%(Med-LM优化提示)至40.5%(GPT-4简单提示)之间,以良性病变更突出。
  • 模型重复性差异较大,从Med-LM优化提示的12.7%波动至GPT-4简单提示的39.0%。

临床应用与阅片要点

  • 阅片报告的结构化程度直接影响人工智能对肝脏病变分类的准确性,优化提示词可部分改善性能。
  • 大语言模型对恶性倾向的正确分类可信度高于完整的LI-RADS分级,但报告中的描述细节缺失会导致分类失败。

局限性

  • 回顾性设计以及5家机构的报告风格差异可能影响模型泛化能力。
  • 两名审阅者间对报告质量的李克特评分一致性仅为弱到中等(Kappa值:0.471-0.766)。
  • 良性病变及无典型恶性描述的病灶的高非分类率与低重复性限制了模型的实用性。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

英文原文(摘要)· 点击展开

BACKGROUND: This study evaluates the performance of a general-purpose (GPT-4) and a medically fine-tuned (Med-LM) large language model (LLM) in classifying liver lesions from unstructured Computed Tomography (CT) reports.

METHODS: Consecutive CT reports (2014-2020) from five institutions were input into GPT-4 and Med-LM withsimple (sp)andoptimised (op) prompts. Lesion-andpatient-level performance were benchmarked against LI-RADS scores assigned by two radiologists, and report quality was analysed using a5-point Likert scale.

RESULTS: A total of 296 CT reports (mean age, 64.6 years ± 11.3 [SD]; 193 men; 654 lesions) were included. Lesion- and patient-level accuracies for LI-RADS scoring ranged from 40.8% (Med-LMsp) to 61.3% (Med-LMop) and from 27.7% (Med-LMsp) to 52.4% (Med-LMop), respectively. When dichotomized into malignant and benign lesions, lesion- and patient-level accuracies rose to 56.1% (GPT-4sp) - 82.3% (Med-LMop) and 71.3% (Med-LMsp) - 86.5% (Med-LMop). Med-LMop demonstrated the highest performance in all analyses and was statistically superior to other models (all p < 0.001). Non-classification rates ranged between 12.7% (Med-LMop) and 40.5% (GPT-4sp), particularly for benign lesions. Kappa values were weak to moderate between the two reviewers in different aspects of report quality (0.471-0.766), and Likert scores for lesion information differed significantly between correctly and incorrectly classified lesions (all p ≤ 0.04). Repeatability varied widely from 12.7% (Med-LMop) to 39.0% (GPT-4sp).

CONCLUSIONS: Med-LM outperforms GPT-4 in classifying liver lesions from unstructured CT reports with both models better at detecting malignancy than full LI-RADS classification. However, high misclassification rates and inconsistent repeatability hinder their clinical use.

来源:Eur J Radiol,查看原文。版权归原作者所有。

引用

  1. Lu, Chiu, Chan, et al. Characterising liver lesions from free-text computer tomography reports - A real-world multicentre analysis. Eur J Radiol. 2026. PMID: 41579671. doi:10.1016/j.ejrad.2026.112689

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。