文献速递神经技术与安全CT

两种人工智能计算机辅助分诊方案检测非增强头颅CT颅内出血的头对头比较

Head-to-Head Comparison of 2 Artificial Intelligence Computer-Aided Triage Solutions for Detecting Intracranial Hemorrhage on Noncontrast Head CT.
PubMed 文献 · AJNR Am J Neuroradiol · 2026-02-03

本研究在单一学术机构内回顾性分析4081例非增强头颅CT,比较两种商用人工智能分诊软件(Vendor A与Vendor B)检测颅内出血的性能。Vendor A敏感度94.4%、特异度97.4%,Vendor B敏感度59.5%、特异度99.0%。研究结论为考虑实施人工智能分诊的临床医生和机构提供了有价值的信息。

要点

  • Vendor A敏感度94.4%、特异度97.4%,Vendor B敏感度59.5%、特异度99.0%。
  • Vendor A漏诊20例,主要为硬膜下和脑实质出血;Vendor B漏诊145例,主要为硬膜下和蛛网膜下腔出血。
  • Vendor A假阳性97例,Vendor B假阳性36例,均与运动伪影等相关。

核心结论

  • 两种人工智能分诊软件在检测颅内出血时性能存在差异,Vendor A敏感度较高而Vendor B特异度较高,结果可为临床医生和医疗机构考虑实施人工智能分诊提供参考。

研究设计与样本

  • 回顾性分析,纳入单一学术机构内多个医疗环境的4081例非增强头颅CT扫描,采用标准化扫描方案,评估两种商用人工智能计算机辅助分诊软件(Vendor A与Vendor B)的病例级分类输出。

诊断标准与关键数值

  • Vendor A:敏感度94.4%,特异度97.4%,阳性预测值77.7%,阴性预测值99.5%。
  • Vendor B:敏感度59.5%,特异度99.0%,阳性预测值85.5%,阴性预测值96.2%。
  • 4081例扫描中595例为颅内出血阳性。

临床应用与阅片要点

  • 作者结论指出,研究结果为临床医生和医疗机构考虑实施人工智能计算机辅助分诊检测颅内出血提供了有价值的信息。
  • 作者讨论提及在缺乏可解释性工具时需结合上下文评估人工智能结果,以及标准化扫描方案对确保人工智能诊断工具可靠性的相关性。

局限性

  • 原文未报告局限性

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 AND PURPOSE: This study aims to provide a comprehensive comparison of the performance and reproducibility of 2 commercially available artificial intelligence (AI) software computer-aided triage and notification solutions, Vendor A (Aidoc) and Vendor B (Viz.ai), for the detection of intracranial hemorrhage (ICH) on noncontrast-enhanced head CT scans performed within a single academic institution.

MATERIALS AND METHODS: The retrospective analysis was conducted on a large patient cohort from multiple health care settings within a single academic institution, utilizing standardized scanning protocols. Sensitivity, specificity, false-positive (FP), and false-negative (FN) rates were evaluated for both vendors. Outputs assessed included AI-generated case-level classification.

RESULTS: Among 4081 scans, 595 were positive for ICH. Vendor A demonstrated a sensitivity of 94.4% and specificity of 97.4%, PPV of 77.7%, and NPV of 99.5%. Vendor B showed a sensitivity of 59.5% and specificity of 99.0%, PPV of 85.5%, and NPV of 96.2%. Vendor A had 20 FNs, which primarily involved subdural and intraparenchymal hemorrhages, and 97 FPs, which appear to be related to motion artifact. Vendor B had 145 FNs, largely comprising of subdural and subarachnoid hemorrhages, and 36 FPs, which appeared to be related to motion artifact and calcified or dense lesions. Concordantly, 18 cases were FNs and 11 cases were FPs for both AI solutions.

CONCLUSIONS: The findings of this study provide valuable information for clinicians and health care institutions considering the implementation of AI software for computer-aided triage and notification in the detection of intracranial hemorrhage. The discussion encompasses the implications of the results, the importance of evaluating AI findings in context-especially in the absence of explainability tools, potential areas for improvement, and the relevance of standardized scanning protocols in ensuring the reliability of AI-based diagnostic tools in clinical practice.

来源:AJNR Am J Neuroradiol,查看原文。版权归原作者所有。

引用

  1. Garcia GM, Young P, Dawood L, et al. Head-to-Head Comparison of 2 Artificial Intelligence Computer-Aided Triage Solutions for Detecting Intracranial Hemorrhage on Noncontrast Head CT. AJNR Am J Neuroradiol. 2026. PMID: 40957692. doi:10.3174/ajnr.A8986

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。