Airway Tree Subtyping on Large Cohorts of CT Images for COPD Risk

针对慢性阻塞性肺病 (COPD) 风险对大组 CT 图像进行气道树亚型分析

• 批准号: 10435540
• 负责人: Andrew Francis Laine
• 金额: $ 65.04万
• 依托单位: COLUMBIA UNIVERSITY HEALTH SCIENCES
• 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 2025-06-30
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10435540
• 关键词: 3-Dimensional; Activities of Daily Living; Air Movements; Asthma; Biological; Caliber; Candidate Disease Gene; Cause of Death; Cessation of life; Chronic Obstructive Pulmonary Disease; Complex; Country; Detection; Development; Developmental Biology; Diagnostic tests; Dictionary; Disease Pathway; Disease susceptibility; Event; Exhibits; Fostering; Funding; General Population; Generations; Genetic; Gold; Grant; Health; High Resolution Computed Tomography; Image; Impairment; Individual; Intervention; Investigation; Knowledge; Label; Link; Longterm Follow-up; Lung; Machine Learning; Manuals; Mathematics; Measures; Minority; Mission; Modeling; Morphology; Multi-Ethnic Study of Atherosclerosis; National Heart, Lung, and Blood Institute; Obstructive Lung Diseases; Occupational; Outcome Measure; Pathway interactions; Phenotype; Prognosis; Public Health; Publishing; Quality of life; Radiology Specialty; Reproducibility; Research; Resolution; Respiratory Signs and Symptoms; Risk; Risk Factors; Role; Scanning; Severities; Shapes; Smoker; Smoking; Spirometry; Standardization; Structure; Supervision; Testing; Textbooks; Time; Tobacco; Trees; United States; United States National Institutes of Health; Variant; Work; X-Ray Computed Tomography; airway obstruction; base; cigarette smoking; clinically relevant; clinically significant; cohort; cost; deep learning; detector; disease prognosis; disorder risk; environmental tobacco smoke; follow-up; genetic epidemiology; genome wide association study; high standard; improved; innovation; lung development; lung lobe; lung volume; machine learning method; never smoker; novel; personalized medicine; pollutant; prognostic significance; pulmonary function decline; respiratory; risk stratification; risk variant; supervised learning; tool; unsupervised learning

Project Summary / Abstract: Chronic obstructive pulmonary disease (COPD) defined by irreversible airflow limitation, is the 3rd leading cause of death globally and 4th in the United States. Smoking tobacco is a major extrinsic COPD risk factor, but despite six decades of declining smoking rates in many countries, the corresponding declines in COPD have been modest. Only a minority of lifetime smokers develop COPD, and up to 25% occurs in never smokers. While other factors have been linked to COPD much of the variation in COPD risk remains unexplained. In addition, personalized risk and therapies are lacking for COPD, due to a lack of reliable COPD subphenotypes. Airflow obstruction, or reduced airflow from the lungs, is determined in part by airway tree structure and lung volume, both of which can be imaged with high precision by high resolution computed tomographic (HRCT) scans. Emerging evidence by our group suggests that airway tree structure variation is common in the general population and is a major contributor to this unexplained COPD risk. By manual labeling of the airway tree structure, limited to one airway generation in just 2 of the 5 lung lobes (due to complexity of tree structure), we found that 26% of the general population has major airway branch variants that differ from the classical “textbook” structure, increase COPD risk, and have a strong and biologically plausible genetic basis. We further demonstrated that airway tree caliber variation (dysanapsis) measured on CT was a stronger predictor of COPD risk than all known risk factors including smoking. Yet there is no standardized approach to characterize the full scope airway tree variation, making the exact relationship between COPD and individual airway-structure features unclear. This proposal would apply for the first-time the power of machine learning methods to the entire airway tree structure imaged on HRCT to build logically upon prior high-impact work to discover new COPD subphenotypes for risk stratification and biological pathways of intervention. Also, we will apply sophisticated / rigorous mathematical clustering approaches to airway trees derived from over 18,000 computed tomography (CT) scans in three highly characterized NIH/NHLBI-funded cohorts – the Multi-Ethnic Study of Atherosclerosis (MESA) Lung Study, the Subpopulations and Intermediate Outcome Measures in Chronic Obstructive Pulmonary Disease Study (SPIROMICS), and the Genetic Epidemiology of COPD (COPDGene) Study, in addition to the Canadian Cohort of Obstructive Lung Disease (CanCOLD) – to discover and replicate novel and clinically significant airway tree subtypes and their genetic basis. The proposed study provides a transformative opportunity to define and validate normal and clinically relevant tree variation in the general population and COPD cohorts. This research would result in robust, reproducible, image based novel quantitative airway tree structure subtypes from lung CT scans, and understand their role in COPD risk, prognosis, and their underlying genetic basis to help personalize COPD risk.

项目概要/摘要： 慢性阻塞性肺疾病（COPD）定义为不可逆的气流限制，是第三大原因 死亡率全球第四，美国第四。吸烟是COPD的主要外在危险因素，但 尽管60年来许多国家的吸烟率下降，但COPD的相应下降 保持谦虚。只有少数终生吸烟者会患上COPD，高达25%的人从不吸烟。 虽然其他因素与COPD有关，但COPD风险的大部分变化仍然无法解释。在 此外，由于缺乏可靠的COPD亚表型，COPD缺乏个性化的风险和治疗。 气流阻塞或来自肺部的气流减少部分由气道树结构和肺 体积，这两者都可以通过高分辨率计算机断层扫描（HRCT）以高精度成像 扫描。我们小组的新证据表明，气道树结构变异在一般情况下是常见的。 这是一个无法解释的COPD风险的主要贡献者。通过手动标记气道树 结构，仅限于在5个肺叶中的2个肺叶中产生一个气道（由于树结构的复杂性）， 我们发现，26%的普通人群的主要气道分支变异不同于经典的 “教科书”结构，增加COPD风险，并具有强大的和生物学上合理的遗传基础。我们进一步 表明CT测量的气道树口径变化（dysanapsis）是COPD的更强预测因子 比所有已知的危险因素，包括吸烟。然而，没有标准化的方法来描述 全方位的气道树变异，使COPD与个体气道结构的关系更加明确 特征不清楚。这一提议将首次将机器学习方法的力量应用于 在HRCT上成像的整个气道树结构在逻辑上建立在以前的高影响力的工作，以发现新的 COPD亚表型用于风险分层和生物学干预途径。 此外，我们将应用复杂/严格的数学聚类方法，从 在三个高度特征化的NIH/NHLBI资助的队列中进行了超过18，000次计算机断层扫描（CT）扫描， 多种族动脉粥样硬化研究（梅萨）肺研究、亚群和中间结局 慢性阻塞性肺疾病研究（SPIROMICS）中的措施，以及 COPD（COPDGene）研究，以及加拿大阻塞性肺疾病队列（CanCOLD）-- 发现和复制新的和临床上重要的气道树亚型及其遗传基础。 拟议的研究提供了一个变革性的机会，以定义和验证正常和临床相关 一般人群和COPD队列中的树变异。这项研究将产生强大的，可重复的， 基于图像的新的定量气道树结构亚型，从肺部CT扫描，并了解他们的作用， COPD风险、预后及其潜在遗传基础，以帮助个性化COPD风险。