A Multiscale Quantitative Comparison of the Normal Lung and Idiopathic Pulmonary Fibrosis

Abstract

The lung functions across multiple spatial scales from the parenchyma, to the blood vessels, and the shape of the lung as it breathes. Pulmonary diseases can affect the lung at any of these spatial scales. Using high resolution computed tomography (HRCT), which is a non-invasive imaging modality, multi-scale spatial information within the lung can be captured. Computer-based quantitative CT (QCT) methods can extract information from HRCT that can be useful for determining disease severity and staging. Idiopathic pulmonary fibrosis (IPF) is a lung disease that permanently and progressively scars the lung. HRCT is an important tool for determining a diagnosis of IPF, however it is very challenging to determine disease trajectory for an individual patient. HRCT-based biomarkers are emerging that might serve this purpose. However, biomarker studies to date only focus on one spatial scale at a time, and a thorough comparison with healthy controls is lacking. This thesis proposes a multi-scale quantitative approach to derive metrics for, quantify, and investigate the pathological features that can be seen on HRCT. By using robust and repeatable QCT methods at different scales, the objective was to quantitatively describe HRCT features that appear during the course of IPF, and to compare these against metrics from an age-matched healthy control cohort. Quadtree decomposition (QtD) was used to quantify heterogeneity for the total lung tissue, and separately for just the radiologically-normal tissue in IPF and controls. Pulmonary vessel-like volume (PVV) was estimated in both cohorts as an index of vascular remodelling, using a graph-based connected-vessel tree analysis. A statistical shape model (SSM) of control (normal) lungs was derived using principal component analysis (PCA), and models for the IPF cohort were derived by projecting to this model. Shape features were compared between cohorts. The potential impact of lung and chest wall shape on the preferential posterior and basal development of fibrosis was studied using a soft tissue mechanics simulation in the normal shape model and an averaged model for the IPF patients. The QtD for the total and the radiologically-normal tissues were, respectively, 55% and 20% higher in IPF than in controls. The finding of increased heterogeneity in the ’normal’ tissue in IPF is a novel outcome with implications for describing disease severity. PVV in IPF was approximately 180% larger than in controls, even when using strict criteria to eliminate all potential artefact. This supports the potential of PVV as a biomarker, but also highlights the need for a standardised approach to its calculation. The first four principal shape modes for the shape model trained to control data were statistically different between controls and IPF, with a distinct clustering for IPF that suggests an IPF shape phenotype. This suggests that there could be a shape phenotype in normal subjects that increases the risk of developing IPF, or that there is a progression towards the IPF shape that could be exploited for staging disease. The mechanics analysis predicted higher tissue elastic recoil pressure in the basal posterior region in the upright lung when the lung shape progressed towards the IPF phenotype. This might have implications for how IPF disease progresses. QtD, PVV and PCA modes all correlated significantly with the extent of fibrosis. In conclusion, IPF appears to have higher heterogeneity in ‘normal’ tissue, a much larger PVV, and quantifiably different shape from normal; together these metrics could serve as repeatable indices for early characterisation of the IPF lung, and for other chronic pulmonary disorders.