文献速递心血管CT

一种用于冠状动脉CT血管成像中斑块定量的全自动深度学习模型

A Fully Automated Deep Learning Model for Quantifying Coronary Plaque at Coronary CT Angiography.
PubMed 文献 · Radiology · 2026-01-01

该研究回顾性纳入17家中国医院2009年6月至2024年5月接受冠状动脉CT血管成像(CCTA)的患者,开发了全自动深度学习斑块定量模型PlaqueSegNet,并在四个外部数据集验证其与血管内超声(IVUS)及专家读者的一致性,同时在三个队列中评估其对主要不良心脏事件的预后价值。

要点

  • PlaqueSegNet在四个外部数据集中定量斑块体积与IVUS及专家读者的一致性良好,组内相关系数均大于0.90。
  • 预测主要不良心脏事件的C指数在中国CT-FFR研究2中为0.64(95%CI:0.62,0.67),在中国CT-FFR研究1.1中为0.65(95%CI:0.60,0.69)。
  • 在连续CCTA队列中C指数为0.74(95%CI:0.66,0.84),提示该模型具有预后价值。

核心结论

  • PlaqueSegNet可基于CCTA全自动测量斑块体积,与专家读者及IVUS高度一致,并对未来主要不良心脏事件具有预后价值。

研究设计与样本

  • 回顾性研究,纳入17家中国医院2009年6月至2024年5月接受CCTA的患者,影像数据按7:3随机分为训练集和验证集以开发全自动深度学习斑块定量模型PlaqueSegNet。

诊断标准与关键数值

  • 训练集1409例(平均年龄63岁±10[SD],795例男性),内部验证集604例(平均年龄63岁±10,329例男性)。

临床应用与阅片要点

  • PlaqueSegNet可全自动定量斑块体积,在四个外部数据集中与IVUS及专家读者的一致性组内相关系数均大于0.90。
  • Bland-Altman分析显示一致性界限较宽。
  • 模型预测主要不良心脏事件的C指数在中国CT-FFR研究2(中位随访2.3年)为0.64(95%CI:0.62,0.67)。
  • 模型预测主要不良心脏事件的C指数在中国CT-FFR研究1.1(中位随访5.3年)为0.65(95%CI:0.60,0.69)。
  • 模型预测主要不良心脏事件的C指数在连续CCTA队列(中位随访3.6年)为0.74(95%CI:0.66,0.84)。

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 Deep learning (DL) models for quantifying plaques at coronary CT angiography (CCTA) are rarely used in routine clinical care. Purpose To develop a fully automated DL model for coronary plaque quantification and to evaluate its prognostic value. Materials and Methods Patients who underwent CCTA were retrospectively enrolled from 17 Chinese hospitals between June 2009 and May 2024. The imaging data of these patients were randomly split into training and validation sets at a 7:3 ratio to develop a fully automated DL model for quantifying plaque volume (PV), PlaqueSegNet, which was subsequently externally tested with four independent datasets: a paired CCTA and intravascular US (IVUS) dataset, a subset of the China CT-derived fractional flow reserve (CT-FFR) study 3 dataset collected with different CT scanners, a serial CCTA dataset within a 3-month interval, and a photon-counting CT dataset. The prognostic value of PlaqueSegNet was evaluated using the Harrell C-index in three cohorts: China CT-FFR study 2, China CT-FFR study 1.1, and a serial CCTA cohort. Results The training dataset included 1409 patients (mean age, 63 years ± 10 [SD]; 795 male), and the internal validation dataset included 604 patients (mean age, 63 years ± 10; 329 male). PlaqueSegNet demonstrated excellent agreement and reproducibility for quantifying PV against IVUS and expert readers across the four external datasets (all intraclass correlation coefficients, >0.90), albeit with wide limits of agreement in Bland-Altman analysis. The C-index of PlaqueSegNet for predicting major adverse cardiac events (MACEs) was 0.64 (95% CI: 0.62, 0.67) in the China CT-FFR study 2 (median follow-up, 2.3 years), 0.65 (95% CI: 0.60, 0.69) in the China CT-FFR study 1.1 (median follow-up, 5.3 years), and 0.74 (95% CI: 0.66, 0.84) in the serial CCTA cohort (median follow-up, 3.6 years). Conclusion PlaqueSegNet provided fully automated measurements of PV from CCTA that closely agreed with expert readers and IVUS and carried prognostic value for future MACEs. Clinical trial registration no. NCT06025305 © RSNA, 2026 Supplemental material is available for this article. See also the editorial by Williams in this issue.

来源:Radiology,查看原文。版权归原作者所有。

引用

  1. Chen Q, Zhou F, Xing W, et al. A Fully Automated Deep Learning Model for Quantifying Coronary Plaque at Coronary CT Angiography. Radiology. 2026. PMID: 42012347. doi:10.1148/radiol.251967

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。