Extended Quantitative Computed Tomography Analysis of Lung Structure and Function

Abstract

Computed tomography (CT) imaging and quantitative CT (QCT) analysis for the study of lung health and disease have been rapidly advanced during the past decades, along with the employment of CT-based computational fluid dynamics (CFD) and machine learning approaches. The work presented in this thesis was devoted to extending the QCT analysis framework from three different perspectives.First, to extend the advanced QCT analysis to more data with undesirably protocolized CT scans, we developed a new deep learning-based automated segmentation of pulmonary lobes, in- corporating z-axis information into the conventional UNet segmentation. The proposed deep learn- ing segmentation, named ZUNet, was successfully applied for QCT analysis of silicosis patients with thick (5 or 10 mm) slices, which used to be excluded in QCT analysis since three-dimensional (3D) volumetric segmentation of the lungs and lobes were hardly successful or not automated. ZUNet outperformed UNet in lobe segmentation of human lungs. In addition, we extended the application of the QCT framework, combining CFD simulations for the entire subjects of the QCT analysis. One-dimensional (1D) CFD simulations of tidal breath- ing have been added to the inspiratory-expiratory CT image matching analysis of 66 asthma pa- tients (M:F=23:43, age=64.4±10.7) for pre- and post-bronchodilator comparison. We aimed to characterize comprehensive airway and lung structure and function relationship in the entire group response and patient-specific response to the bronchodilator. Along with the evidence of large air- way dilatation in the entire asthmatics, the CFD analysis revealed that improvements in regional flow rate fraction, particularly in the right lower lobe (RLL), airway pressure drop, airway resis- tance, and workload of breathing were significantly associated with the degree of large airway dilatation. Finally, we extended the approach using machine learning analysis to integrate numerous QCT variables with clinical features and additional information such as environmental exposure. In pursuit of investigating the effects of particulate matter (PM) exposure on human lung struc- ture and function alteration, principal component analysis (PCA) and k-means clustering iden- tified low, mid, and high exposure groups from directly measured air pollution exposure data of 270 healthy (age=68±10, M:F=15:51), asthma (age=60±12, M:F=39:56), chronic obstructive pulmonary disease (COPD) (age=69±7, M:F=66:10), and idiopathic pulmonary fibrosis (IPF) (age=72±7, M:F=43:10) subjects. Based on the exposure clusters, the RLL segmental airway narrowing was observed in the high exposure group. Various associations were found between the exposure data and about 200 multiscale lung features, from quantitative inspiratory and ex- piratory CT image matching and 1D CFD tidal breathing simulations. To highlight, small PM increases small airway disease in asthma. PM at all sizes decreases inspiratory low attenuation area in COPD and diseases luminal diameter of the RLL segmental airways in IPF.