文献速递神经技术与安全CT

人工智能自动筛选神经放射学光子计数CT适用患者

Automatic Patient Eligibility for Photon-counting CT Using Discriminative and Generative AI Models in Neuroradiology.
PubMed 文献 · Acad Radiol · 2026-08-05

本研究回顾性纳入800份西班牙语神经放射学CT申请,由两名神经放射科医生标注为基础方案或高级方案,以专家共识为金标准。比较判别式Transformer模型与生成式大语言模型的分类性能。结果显示判别式模型RoBERTa表现最佳,准确率0.925,显著优于生成式大语言模型,可用于自动化PCCT适用性分流。

要点

  • 判别式Transformer模型RoBERTa可自动识别需行PCCT高级方案的CT申请,准确率达0.925。
  • 通用生成式大语言模型(ChatGPT、Gemini)在PCCT适用性分类中表现最差,不应直接用于临床分流。
  • 西班牙语BERT与RoBERTa分类性能无显著差异,均可作为PCCT自动分流候选模型。

核心结论

  • 判别式Transformer模型,尤其是RoBERTa,可高精度自动识别需PCCT高级方案的CT申请,显著优于通用生成式大语言模型。
  • 本研究为探索性概念验证,支持AI驱动的PCCT适用性自动分流以优化利用并精简神经放射学工作流程。

研究设计与样本

  • 随机抽取800份申请,由两名神经放射科医生独立标注为基础方案或高级方案,专家共识为金标准,Cohen's k = 0.74。
  • 判别式模型采用西班牙语BERT、生物医学RoBERTa、Llama-3.1-8B和Mistral-7B进行微调,以分层五折交叉验证评估。
  • 生成式模型ChatGPT 5.2和Gemini 3在零样本设置下使用结构化提示,后处理为二分类输出。

诊断标准与关键数值

  • 基础方案:可在EID-CT或基础PCCT上完成;高级方案:需要全功能高级PCCT完成。
  • 性能评估指标包括准确率、精确率、召回率、特异度、F1值和AUC。
  • RoBERTa表现最佳:准确率0.925,精确率0.906,召回率0.967,F1 0.935,AUC 0.920,总误差最低。
  • RoBERTa与西班牙语BERT之间无显著差异;RoBERTa显著优于Llama、Mistral、Gemini和ChatGPT。
  • 生成式大语言模型整体性能最差。

临床应用与阅片要点

  • 可考虑将判别式Transformer模型集成到CT预约流程中,自动将申请分流至PCCT或EID-CT。
  • RoBERTa的高召回率(0.967)提示漏检需PCCT高级方案的可能性较低,适合作为筛查工具。
  • 基础方案与高级方案的二分法可作为神经放射学PCCT适用性的初步分类框架。
  • 当前无指南明确规定神经放射学中何时应优先选择PCCT而非EID-CT,AI分流可部分弥补这一空白。

局限性

  • 本研究为单中心回顾性探索性概念验证研究。
  • 协同一致性Cohen's k为0.74,提示标注一致性虽可接受但并非极高。
  • 仅使用西班牙语文本,跨语言泛化性尚未验证。
  • 生成式大语言模型为零样本设置,未经特定任务微调。

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

RATIONALE AND OBJECTIVES: Photon -Counting CT (PCCT) provides higher resolution, reduced dose, and valuable spectral data, but generates a large number of images per study, underlining radiologist and Picture Archiving and Communication Systems (PACS) capacity. While no guidelines defining when PCCT should be preferred over conventional Energy Integrating Detector (EID) CT in neuroradiology, and manual decision-making being impractical, Natural Language Processing (NLP)-based large language models (LLMs) may help automate routing of CT requests to the most appropriate CT technology.

MATERIALS AND METHODS: We conducted a retrospective study using Spanish-language neuroradiology CT requests retrieved from the Radiology Information System (RIS) (January 2012-October 2025). A random sample of 800 requests was independently labeled by two neuroradiologists as Basic Protocol (BP); feasible on EID-CT or basic PCCT) or Advanced Protocol (AP); requiring full-advanced PCCT), considering the gold standard an expert consensus (Cohen's k = 0.74). For discriminative modeling, transformer classifiers (Spanish bidirectional encoder representations from transformers (BERT), biomedical RoBERTa, Llama-3.1-8B, and Mistral-7B) were fine-tuned for binary BP/AP classification and evaluated with stratified five-fold cross-validation. For generative modeling, ChatGPT 5.2 and Gemini 3 were applied in a zero-shot setting using structured prompting and post-processed into binary outputs. Performance was assessed using accuracy, precision, recall, specificity, F1-score, and AUC.

RESULTS: Discriminative transformers achieved the best performance for the classification, with RoBERTa leading (accuracy 0.925, precision 0.906, recall 0.967, F1 0.935, AUC 0.920) and the lowest total error, without a significant difference versus Spanish BERT. In contrast, RoBERTa outperformed Llama, Mistral, Gemini, and ChatGPT with statistically significant improvements, while the generative LLMs yielded the poorest overall performance.

CONCLUSION: Discriminative transformer models, particularly RoBERTa, enabled highly accurate automatic identification of CT requests requiring AP to be performed at PCCT, substantially outperforming general-purpose generative LLMs. These findings support this exploratory proof-of-concept study of AI-driven eligibility routing to optimize PCCT utilization and streamline neuroradiology workflows.

来源:Acad Radiol,查看原文。版权归原作者所有。

引用

  1. Martín-Noguerol, López-Úbeda, Escartín, et al. Automatic Patient Eligibility for Photon-counting CT Using Discriminative and Generative AI Models in Neuroradiology. Acad Radiol. 2026. PMID: 42557204. doi:10.1016/j.acra.2026.07.026

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。