Department of Radiology and Radiological Science, Division of Cardiovascular Imaging, Medical University of South Carolina, Charleston, SC.
Institute of Clinical Radiology and Nuclear Medicine, University Medical Center Mannheim, Medical Faculty Mannheim, Heidelberg University, Heidelberg.
J Thorac Imaging. 2020 May;35 Suppl 1:S28-S34. doi: 10.1097/RTI.0000000000000500.
The objective of this study was to evaluate an artificial intelligence (AI)-based prototype algorithm for the fully automated per lobe segmentation and emphysema quantification (EQ) on chest-computed tomography as it compares to the Global Initiative for Chronic Obstructive Lung Disease (GOLD) severity classification of chronic obstructive pulmonary disease (COPD) patients.
Patients (n=137) who underwent chest-computed tomography acquisition and spirometry within 6 months were retrospectively included in this Institutional Review Board-approved and Health Insurance Portability and Accountability Act-compliant study. Patient-specific spirometry data, which included forced expiratory volume in 1 second, forced vital capacity, and the forced expiratory volume in 1 second/forced vital capacity ratio (Tiffeneau-Index), were used to assign patients to their respective GOLD stage I to IV. Lung lobe segmentation was carried out using AI-RAD Companion software prototype (Siemens Healthineers), a deep convolution image-to-image network and emphysema was quantified in each lung lobe to detect the low attenuation volume.
A strong correlation between the whole-lung-EQ and the GOLD stages was found (ρ=0.88, P<0.0001). The most significant correlation was noted in the left upper lobe (ρ=0.85, P<0.0001), and the weakest in the left lower lobe (ρ=0.72, P<0.0001) and right middle lobe (ρ=0.72, P<0.0001).
AI-based per lobe segmentation and its EQ demonstrate a very strong correlation with the GOLD severity stages of COPD patients. Furthermore, the low attenuation volume of the left upper lobe not only showed the strongest correlation to GOLD severity but was also able to most clearly distinguish mild and moderate forms of COPD. This is particularly relevant due to the fact that early disease processes often elude conventional pulmonary function diagnostics. Earlier detection of COPD is a crucial element for positively altering the course of disease progression through various therapeutic options.
本研究旨在评估一种基于人工智能（AI）的原型算法，用于在胸部 CT 上实现全自动化的肺叶分割和肺气肿定量（EQ），并将其与慢性阻塞性肺疾病（COPD）患者的全球倡议慢性阻塞性肺病（GOLD）严重程度分类进行比较。
本回顾性研究纳入了在 6 个月内接受胸部 CT 采集和肺功能检查的患者。患者的特定肺功能数据，包括 1 秒用力呼气量（FEV1）、用力肺活量（FVC）和 1 秒用力呼气量/用力肺活量比值（Tiffeneau 指数），用于将患者分配到各自的 GOLD 1 至 4 期。使用 AI-RAD Companion 软件原型（西门子医疗）进行肺叶分割，该软件原型是一种深度卷积图像到图像的网络，并在每个肺叶中定量肺气肿以检测低衰减体积。
全肺-EQ 与 GOLD 分期之间存在很强的相关性（ρ=0.88，P<0.0001）。相关性最强的是左肺上叶（ρ=0.85，P<0.0001），最弱的是左肺下叶（ρ=0.72，P<0.0001）和右中叶（ρ=0.72，P<0.0001）。
基于 AI 的肺叶分割及其 EQ 与 COPD 患者的 GOLD 严重程度分期具有非常强的相关性。此外，左肺上叶的低衰减体积不仅与 GOLD 严重程度相关性最强，而且还能够最清楚地区分 COPD 的轻度和中度形式。这一点尤其重要，因为早期疾病过程通常逃避常规肺功能诊断。早期发现 COPD 对于通过各种治疗选择积极改变疾病进展的进程至关重要。
