文献速递心血管胸部CT

人工智能标准化冠周脂肪衰减指数测量:多厂商CT外部验证

Artificial intelligence standardizes pericoronary fat attenuation index measurement: external multi-vendor CT validation.
PubMed 文献 · Br J Radiol · 2026-01-01

本研究回顾性纳入1125例患者共3375支冠状动脉,使用5种不同厂商、型号和代次的CT设备评估AI自动化测量冠周脂肪衰减指数(FAI)的可行性与泛化性。结果显示AI-FAI与工作站FAI高度一致,诊断性能相当(AUC=0.851),且跨5种扫描仪稳定性良好,为心血管风险评估提供了可靠的自动化工具。

要点

  • AI-FAI与工作站FAI一致性良好(ρ=0.877,平均偏倚1.60 HU),日常报告可参考自动化测量结果。
  • 以-70.1 HU为阈值时,AI-FAI诊断准确率84.98%,AUC=0.851,与工作站性能相当。
  • AI-FAI跨5种主流CT设备表现稳定,扫描仪间ρ为0.829至0.896,利于多院区标准化应用。

核心结论

  • AI-FAI与工作站FAI一致性良好、诊断性能可靠、跨扫描仪稳健,可作为心血管风险评估的自动化替代工具。

研究设计与样本

  • 回顾性分析1125例患者(男性584例,占51.9%;平均年龄61.5±12.3岁;共3375支冠状动脉)的冠状动脉CT血管造影数据。
  • 使用5种不同厂商、型号和代次的主流CT扫描仪获取数据,分别以AI软件和工作站定量评价主要冠状动脉周围FAI值。
  • 评估方法间一致性、诊断性能及扫描仪对测量结果的影响。

关键影像表现/征象

  • AI-FAI与工作站FAI在患者水平呈良好相关(ρ=0.877,P<.001;n=1125),平均偏倚1.60 HU。
  • 以-70.1 HU为阈值时,AI-FAI准确率为84.98%,诊断性能AUC=0.851,与工作站FAI相当。
  • AI-FAI在全部5种扫描仪上表现稳定,扫描仪间ρ范围为0.829至0.896。
  • 超出一致性界限的离群值比例仅为3.70%至6.06%。

诊断标准与关键数值

  • 以冠周脂肪衰减指数阈值-70.1 HU作为判定标准时,AI-FAI诊断准确率为84.98%,AUC=0.851。
  • AI-FAI与工作站FAI一致性界限为95%一致性界限-4.94至8.14 HU,94.84%的测量值落在界限内。

临床应用与阅片要点

  • AI-FAI可作为工作站FAI的实用替代方案,适用于临床常规冠脉CTA阅片中的冠周炎症定量。
  • 跨多厂商CT的稳定性支持在不同设备间推广AI-FAI标准化测量,减少手动操作变异。
  • 临床报告中可参考AI自动生成的FAI值辅助心血管风险评估,注意与工作站测量值间约1.60 HU的平均偏倚。

局限性

  • 本研究为回顾性分析,未提供前瞻性队列验证结果。
  • 原文未报告外部临床结局随访数据,FAI与心血管事件预后的关联未在本研究中评估。

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

OBJECTIVES: To evaluate the feasibility and generalizability of an AI-based automated method for measuring the pericoronary fat attenuation index (FAI)-a promising non-invasive imaging biomarker for coronary inflammation-and to promote efficient clinical translation.

METHODS: A retrospective analysis was conducted on coronary computed tomography angiography data from 1125 patients (584 [51.9%] male; mean age 61.5 ± 12.3 years; totaling 3375 coronary arteries) examined using 5 mainstream CT scanners from different manufacturers, models, and generations. FAI values around the major coronary arteries were quantitatively evaluated using both AI software and workstation. Agreement between methods, diagnostic performance, and scanner influence were assessed.

RESULTS: Artificial intelligence-based fat attenuation index (AI-FAI) demonstrated a favorable correlation with workstation-FAI at the patient level (ρ = 0.877, P < .001; n = 1125), along with a high level of agreement (mean bias: 1.60 HU; 95% limits of agreement: -4.94 to 8.14; 94.84% within limits). Using a threshold of -70.1 HU, AI-FAI demonstrated performance comparable to workstation-FAI, with accuracy (84.98%) and solid diagnostic performance (AUC = 0.851). Furthermore, AI-FAI exhibited stable measurement performance across all 5 scanners, with inter-scanner ρ ranging from 0.829 to 0.896 and only 3.70%-6.06% of outliers beyond the agreement limits.

CONCLUSIONS: AI-FAI demonstrates favorable agreement with traditional workstation-FAI, solid diagnostic performance, and consistent cross-scanner robustness, providing a reliable automated tool for cardiovascular risk assessment.

ADVANCES IN KNOWLEDGE: This is a novel study demonstrating AI-FAI as a practical alternative to workstation-FAI across multiple CT scanners, facilitating reliable quantification of coronary inflammation for routine clinical practice.

来源:Br J Radiol,查看原文。版权归原作者所有。

引用

  1. Qiu, Wu, Zhao, et al. Artificial intelligence standardizes pericoronary fat attenuation index measurement: external multi-vendor CT validation. Br J Radiol. 2026. PMID: 42606570. doi:10.1093/bjr/tqag186

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。