Department of Thoracic Imaging, Massachusetts General Hospital, 75 Blossom Court, Boston, MA, USA.
Harvard Medical School, Boston, MA, USA.
Eur Radiol. 2020 May;30(5):2535-2542. doi: 10.1007/s00330-019-06607-9. Epub 2020 Jan 31.
To assess quantitative lobar pulmonary perfusion on DECT-PA in patients with and without pulmonary embolism (PE).
Our retrospective study included 88 adult patients (mean age 56 ± 19 years; 38 men, 50 women) who underwent DECT-PA (40 PE present; 48 PE absent) on a 384-slice, third-generation, dual-source CT. All DECT-PA examinations were reviewed to record the presence and location of occlusive and non-occlusive PE. Transverse thin (1 mm) DECT images (80/150 kV) were de-identified and exported offline for processing on a stand-alone deep learning-based prototype for automatic lung lobe segmentation and to obtain the mean attenuation numbers (in HU), contrast amount (in mg), and normalized iodine concentration per lung and lobe. The zonal volumes and mean enhancement were obtained from the Lung Analysis™ application. Data were analyzed with receiver operating characteristics (ROC) and analysis of variance (ANOVA).
The automatic lung lobe segmentation was accurate in all DECT-PA (88; 100%). Both lobar and zonal perfusions were significantly lower in patients with PE compared with those without PE (p < 0.0001). The mean attenuation numbers, contrast amounts, and normalized iodine concentrations in different lobes were significantly lower in the patients with PE compared with those in the patients without PE (AUC 0.70-0.78; p < 0.0001). Patients with occlusive PE had significantly lower quantitative perfusion compared with those without occlusive PE (p < 0.0001).
The deep learning-based prototype enables accurate lung lobe segmentation and assessment of quantitative lobar perfusion from DECT-PA.
• Deep learning-based prototype enables accurate lung lobe segmentation and assessment of quantitative lobar perfusion from DECT-PA. • Quantitative lobar perfusion parameters (AUC up to 0.78) have a higher predicting presence of PE on DECT-PA examinations compared with the zonal perfusion parameters (AUC up to 0.72). • The lobar-normalized iodine concentration has the highest AUC for both presence of PE and for differentiating occlusive and non-occlusive PE.
评估 CT 肺动脉造影（CTPA）双能量去卷积技术（DECT）中定量肺叶灌注在有和无肺栓塞（PE）患者中的应用。
本回顾性研究纳入 88 例成年患者（平均年龄 56±19 岁；男 38 例，女 50 例），均行 384 层第三代双源 CT 的 CTPA 检查（40 例有 PE，48 例无 PE）。所有 CTPA 检查均进行评估，记录是否存在并定位闭塞性和非闭塞性 PE。对横断位（1mm）DECT 图像（80/150kV）进行去识别处理，并离线导入到独立的深度学习基础原型中，以进行自动肺叶分割并获得平均衰减值（HU）、对比量（mg）、肺和肺叶的标准化碘浓度。区域性容积和平均增强值从 Lung Analysis™应用程序中获得。采用受试者工作特征（ROC）和方差分析（ANOVA）进行数据分析。
DECT-PA 的自动肺叶分割在所有患者中均准确（88 例；100%）。与无 PE 患者相比，有 PE 患者的肺叶和区域性灌注均明显降低（p<0.0001）。与无 PE 患者相比，有 PE 患者的各肺叶平均衰减值、对比量和标准化碘浓度均显著降低（AUC 0.70-0.78；p<0.0001）。与无闭塞性 PE 患者相比，闭塞性 PE 患者的定量灌注明显降低（p<0.0001）。
基于深度学习的原型可准确进行肺叶分割，并从 CTPA 评估定量肺叶灌注。
基于深度学习的原型可准确进行肺叶分割，并从 CTPA 评估定量肺叶灌注。
与区域性灌注参数（AUC 最高 0.72）相比，定量肺叶灌注参数（AUC 最高 0.78）在预测 CTPA 检查中的 PE 存在方面具有更高的预测价值。
对于 PE 的存在以及区分闭塞性和非闭塞性 PE，肺叶标准化碘浓度的 AUC 最高。
