Analysis of Airways in Computed Tomography

Computed Tomography (CT) allows un-occluded three-dimensional views of the interior of scanned objects. In medical applications this offers the opportunity to diagnose or study diseases, such as Chronic Obstructive Pulmonary Disease (COPD) with limited consequence to the patients. COPD is major cause of death and disability world-wide. It affects how lungs function through two competing mechanisms: destruction of lung tissue known as emphysema and inflammation of airways, leading to thickened airway walls and narrowed airway lumen. Diagnosis of COPD is based on airflow limitation as determined with lung function tests, but CT adds to our understanding of disease mechanisms by for instance enabling lung density measurements, which gives insight into emphysema distribution within the lungs. The airways with their tree-like structure have so far been more challenging to analyse. Automated methods are indispensable as the visible airway tree in a CT scan can include several hundreds of individual branches. Automation of measurements is difficult, however, due to the airway’s complicated structure, which varies in size and shape; biologically between subjects and dynamically during breathing. This thesis presents several methods for solving problems related to analysing airways and results of using these methods to study COPD via data from the Danish Lung Cancer Screening Trial. A first step in analysing structures from images, is segmentation. A graph based approach for this is presented, which is able to simultaneously find the inner and outer airway wall surfaces in three dimensions. Measurements of airway wall surfaces depend on the branch they are made in. A method is presented that is able to match branches in multiple scans of the same subject using image registration, allowing measurements to be followed within single branches over time. Between subjects, measurements can be compared if they are done in the same anatomical subset of branches. A supervised algorithm for anatomical labelling of branches based on geodesic distances is presented, allowing such comparisons. Branch matching only allow changes from scan to scan to be measured if the branches are successfully segmented in each scan. To improve on this, the segmentation approach is extended to allow multiple scans to be segmented simultaneously by combining information from all images. The resulting surfaces correspond, meaning changes can be measured locally without the need for a separate branch matching step. The developed fully automatic framework is applied to study effect of differences in inspiration level at the time of scan on airway dimensions in subjects with and without COPD. Results show measured airway dimensions, in scans close to maximum inspiration, to be affected by un-intended differences in the level of inspiration and this dependency is again influenced by COPD. Inspiration level should therefore be accounted for when measuring airways, and airway abnormalities typically associated with airflow limitation, such as airway wall thickening and lumen narrowing, should at least partly be understood as being due to differences in how airways are influenced by the inspiration level in subjects with and without COPD.