Deep Learning and Fluid Dynamics Based Phenotyping of Expiratory Central Airway Collapse

点击翻译按钮获取中文摘要

Project Summary/ Abstract Expiratory central airway collapse (ECAC), defined by >50% collapse of large airways during expiration, resulting from either cartilaginous weakening or redundancy of the posterior membranous wall of the trachea, is an increasingly recognized disorder associated with cigarette smoking and chronic obstructive pulmonary disease (COPD). Airflow obstruction in smokers primarily arises from increased resistance to airflow in the small distal conducting airways <2 mm in diameter. It is plausible that in a subset of smokers with and without COPD, central airway collapse results in additional resistance to airflow, resulting in substantial respiratory morbidity. Ninety-two million adults in the Unites States are active or past smokers, and ECAC is present in approximately 5% of current and former smokers. The presence of ECAC is associated with greater dyspnea, worse respiratory-quality of life and greater frequency of exacerbations after adjustment for underlying lung disease. Whether these patients will benefit from interventional therapies such as stenting or tracheopexy depends on whether the airflow resistance caused by ECAC contributes to symptoms, and this in turn depends on the relative contribution of central and small airways to overall airflow resistance. If the overall airflow resistance is primarily due to distal small airways obstruction in a given patient with ECAC, treating central airway collapse is unlikely to benefit such a patient. Our central hypothesis is that ECAC results in additional airflow obstruction beyond that incurred in the small airways, and that in a subset of patients the central airways are the major site of airflow obstruction and hence are amenable to therapy. The complex interplay of proximal and distal airway resistances and transpulmonary pressures does not lend itself to direct measurements in human subjects across a range of physiological pressure and flow changes. We propose a combination of CT-derived imaging and patient-personalized benchtop model and deep learning to answer these questions with the following specific aims. Aim 1 of this application will be to derive personalized patient- specific information on airway geometry and resistance using airway segmentation from computed tomography (CT) scans. We will calculate airway resistances in central and small airways using standard formulae. The goal of Aim 2 is to create bench-top simulations to understand the complex interplay between the resistance of small and large airways. In Aim 3, we will use deep learning to derive probability scores for clinically substantial ECAC from segmented airway images on computed tomography. The results of our study will enable patient-specific personalized therapies for ECAC. The mechanistic insights gained from this study will help identify patients with clinically significant ECAC and hence most likely to benefit from therapeutic interventions.