Prognostic Markers of Emphysema Progression

项目总结/摘要 慢性阻塞性肺疾病（COPD）影响美国多达2400万人， 预计到2020年将成为全球第三大死亡原因，总成本为500亿美元。COPD有 传统上被分为肺气肿和慢性支气管炎的临床表型，但其 对潜在的机制知之甚少。特别是，肺气肿被定义为异常的，永久性的， 远端空气空间的扩张。这一病理过程的发展和进展与 伴随肺功能下降和进行性临床损害。计算机断层扫描（CT）成像 胸部越来越多地被用来客观地量化疾病及其进展。电流 量化肺气肿进展的方法是有限的，并且丢弃了大部分空间和时间的 在吸气和呼气时获得的CT扫描信息。在这份提案中，我们计划开发 基于图像密度标记的计算组件，用于预测肺气肿进展 和肺机械应变特性取决于其潜在的肺气肿亚型。这项建议 利用我们以前在计算肺气肿亚型方面的经验， 围绕肺气肿进展的假设机制量身定制的一组新型预后标志物： 炎症损伤和机械性劳损。为了达到这个目标，我们将（1）发展晚期肺气肿 使用新型深度学习架构的子类型方法，（2）开发一种快速的质量保持大型 位移配准方法，以使得能够发现肺组织的局部弹性特性， 吸气和呼气CT扫描，（3）基于图像发现新亚型特异性生物标志物特征 密度关系和机械性能使用无监督深度学习技术在一个共同的 统计框架，和（4）验证所提出的生物标志物的预后价值及其关联 降低终点和临床结果，以实现其临床解释和翻译。除了 将探索基于先进机器学习技术的替代预后模型， 进行了一项模型比较研究，以确定肺气肿进展的最佳预后模型。我们 分析将处理对应于5，517名受试者的12，300次扫描，基线和随访数据来自 COPDGene队列-COPD中最大的队列之一，包含吸气和呼气时的CT图像， 呼吸和遗传测量。所提出的方法将提供可重复的，自动的和 低成本的肺气肿进展的体内预后生物标志物， 并将其转化为临床实践。

Project Summary/Abstract Chronic Obstructive Pulmonary Disease (COPD) affects up to 24 million people in the United States and is projected to be the 3rd leading cause of death worldwide by 2020 with a total cost of $50 billion. COPD has been traditionally dichotomized into the clinical phenotypes of emphysema and chronic bronchitis, but its underlying mechanisms are poorly understood. In particular, emphysema is defined as abnormal, permanent dilation of the distal airspaces. The development and progression of this pathologic process are associated with a decline in lung function and progressive clinical impairment. Computed tomographic (CT) imaging of the chest is increasingly being leveraged to quantify the disease and its progression objectively. Current approaches to quantify emphysema progression are limited and discard most of the spatial and temporal information in CT scans obtained at inspiration and expiration. In this proposal, we plan on developing computational components to prognosticate emphysema progression that builds upon image density markers and lung mechanical strain characteristics conditioned on their underlying emphysema subtypes. This proposal leverages our previous experience in computational emphysema subtyping to discover, validate and translate a novel panel of prognostic markers tailored around the postulated mechanisms of emphysema progression: inflammation injury and mechanical strain. To reach this goals, we will (1) develop an advanced emphysema subtyping approach using novel deep learning architectures, (2) develop a fast mass preserving large displacement registration approach to enable the discovery of local elastic properties of lung tissue between inspiratory and expiration CT scans, (3) discover new subtype-specific biomarker features based on image density relations and mechanical properties using unsupervised deep learning techniques within a common statistical framework, and (4) validate the prognostic value of the proposed biomarkers and their association with decline end-points and clinical outcomes to enable its clinical interpretation and translation. In addition to that, will be explored alternative prognostic models based on advanced machine learning techniques and performed a model comparison study to define the most prognostic model for emphysema progression. Our analysis will process 12,300 scans corresponding to 5,517 subjects with baseline and follow-up data from the COPDGene cohort –one of the largest cohort in COPD containing CT images at inspiration and expiration, respiratory and genetic measurements. The proposed methodology will provide reproducible, automatic and low-cost prognostic in-vivo biomarkers of emphysema progression that may enable the discovery of new therapies and translate them into clinical practice.