Accuracy and Precision in CT Quantification of COPD Through Virtual Imaging Trials

通过虚拟成像试验对 COPD 进行 CT 定量的准确性和精确度

• 批准号: 10435577
• 负责人: Ehsan Abadi
• 金额: $ 44.76万
• 依托单位: DUKE UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-07-01 至 2026-06-30
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10435577
• 关键词: Algorithms; Anatomy; Artificial Intelligence; Biological Markers; Cause of Death; Chronic Obstructive Pulmonary Disease; Communities; Comparative Study; Complement; Complex; Computed Tomography Scanners; Computer Models; Coupled; Data; Data Set; Densitometry; Diagnosis; Diagnostic; Diagnostic Procedure; Disease; Disease model; Effectiveness; Environmental Risk Factor; Evaluation; Exposure to; Goals; Image; Interstitial Lung Diseases; Libraries; Longitudinal Studies; Lung; Lung diseases; Magnetic Resonance Imaging; Measurement; Methods; Modality; Modeling; Monitor; Patient imaging; Patients; Persons; Photons; Physiology; Positioning Attribute; Prevalence; Protocols documentation; Provider; Pulmonary Emphysema; Radiation exposure; Recording of previous events; Reporting; Reproducibility; Research; Resolution; Role; Severities; Severity of illness; Smoking; Spirometry; Symptoms; System; Technology; Time; Tube; Variant; X-Ray Computed Tomography; artificial intelligence algorithm; base; cohort; disease diagnosis; human model; image processing; imaging biomarker; improved; in silico; in vivo; insight; quantitative imaging; radiation absorbed dose; reconstruction; tool; virtual; virtual imaging; virtual patient; voltage

Chronic Obstructive Pulmonary Disease (COPD) is a leading cause of death. Increasing in prevalence, COPD is a major burden to patients and providers. Computed tomography (CT) can provide valuable information about the structural and functional abnormalities of the disease as demonstrated in numerous studies where quantitative CT is deployed to characterize and evaluate the treatment. For instance, the COPDGene study has recently shown the substantial role of quantitative CT in the redefinition of COPD diagnosis, and in evaluating the progression of emphysema over time. However, these biomarkers vary across different scanners, settings, and patient attributes. There is a crucial need to manage this variability by optimizing and harmonizing CT images for reliable biomarker quantifications across both current and emerging scanners. This goal is not possible through conventional methods of using physical phantoms or patient images. Physical phantoms are often oversimplified and not representative of the complex anatomy and physiology of COPD patients. Patient images are ground-truth-limited, i.e., the exact anatomy and physiology of the patient is not fully known. Further, patient-based comparisons require multiple acquisitions of the same subjects across different scanners and settings. This is not ethically possible since repeated imaging increases the absorbed radiation dose. These challenges can be overcome through the use of virtual imaging trials (VITs) where studies are performed in silico using computational models of patients and scanners. VITs can provide reliable and practical solution to the challenge of COPD imaging provided realistic models of patients and scanners. Such models are currently lacking in the context of COPD. We develop and then utilize realistic virtual imaging toolsets to systematically evaluate and optimize CT biomarkers in COPD patients across scanners, imaging parameters, and patient attributes. We develop the first library of realistic COPD patient models with diverse attributes and severities. Coupled with accurate models of different scanners, the phantoms will be used to generate sets of ground-truth-known virtual CT cases, to be disseminated to the research community and to be used to systematically evaluate the effects of current and emerging scanners, various patient attributes, and the effects of image processing algorithms (through a national challenge), on the accuracy and precision of COPD biomarkers. Further, we develop and optimize a truth-based artificial intelligence-based algorithm for COPD quantifications. We optimize the algorithm for accuracy and reproducibility, taking advantage of the ground-truth known simulated images . We then harmonize CT settings across different scanners to accurately and precisely assess COPD imaging biomarkers for both single time-point and longitudinal studies. The studies will be done for the top two image processing algorithms, identified in the challenge, as well as our developed algorithm. Through these efforts, the project will position CT as a more reliable method for improved characterization and monitoring of COPD.

慢性阻塞性肺疾病(COPD)是导致死亡的主要原因。慢性阻塞性肺病患病率上升 是患者和提供者的主要负担。计算机断层扫描(CT)可以提供有价值的信息 关于疾病的结构和功能异常，如许多研究所证明的那样， 定量CT被用来描述和评估治疗。例如，COPD基因研究 最近显示了定量CT在重新定义COPD诊断中的重要作用，以及在 评估肺气肿随时间的发展。然而，这些生物标志物在不同的 扫描仪、设置和患者属性。迫切需要通过优化和 协调CT图像，以便在现有和新兴扫描仪上进行可靠的生物标记物量化。 通过使用物理模型或患者图像的传统方法，这一目标是不可能实现的。物理 幻影往往过于简单化，不能代表COPD的复杂解剖和生理学 病人。患者的图像是受事实限制的，即患者的确切解剖和生理学不是 完全出名了。此外，以患者为基础的比较需要多次获取相同的受试者 不同的扫描仪和设置。这在伦理上是不可能的，因为重复成像会增加吸收的 辐射剂量。这些挑战可以通过使用虚拟成像试验(VIT)来克服 研究使用患者和扫描仪的计算模型在电子计算机中进行。VITS可以提供可靠的 而实际解决COPD影像的挑战，为患者和扫描仪提供了逼真的模型。 在慢性阻塞性肺病的背景下，目前缺乏这样的模型。 我们开发并利用逼真的虚拟成像工具集来系统地评估和优化CT 不同扫描仪、成像参数和患者属性的COPD患者的生物标记物。我们开发了 第一个具有不同属性和严重程度的现实COPD患者模型库。再加上准确的 不同型号的扫描仪，这些模型将被用来生成一组已知地面真相的虚拟CT 案例，向研究界传播，并用于系统地评估 当前和新兴的扫描仪、各种患者属性以及图像处理算法的效果 (通过国家挑战)，关于COPD生物标志物的准确性和精密度。此外，我们还开发和 优化基于真理的基于人工智能的COPD量化算法。我们优化了 利用地面实况已知模拟图像的准确性和重复性的算法 。我们 然后协调不同扫描仪的CT设置，以准确和精确地评估COPD成像 单一时间点和纵向研究的生物标记物。 研究将针对排名前两位的图像进行 处理算法，在挑战中确定，以及我们开发的算法。通过这些努力， 该项目将使CT成为一种更可靠的方法，用于改善COPD的特征和监测。