Abstract
This thesis presents a solution for comparing and identifying air trapping profiles among asthmatic patients. Asthma is defined as a narrowing of the airways following irritation. The mechanical reactions associated with irritation affect the expiratory flow by reducing the bronchial lumen, and thus causing exacerbation. This air flow reduction leads to the non-evacuation of air from certain regions of the lung upon expiration, a phenomenon named air trapping. Though triggers are well known, the mechanisms underlying the irritation are complex and poorly understood, the lung being a complex internal organ. This work is primarily intended to detect and describe different trapping profiles which correspond to specific sensitivities of the bronchial tree between patients. We also sought to associate a clinical data with trapping profiles. This work is based on data from patient scans collected during a bronchial reactivity test designed to capture the evolution of airway obstruction. The statistical models consists of comparing trapping profiles in a standardized way. Such profiles are derived by isolating pulmonary parenchyma on CT images, then generating a data space with mathematical properties enabling analysis. In this space, trapping profiles are characterized by the distribution of their nearest neighbors. This makes it possible to obtain for each image a local representation of the trapping distribution. The estimator of this distribution is standardized by a theoretical uniform distribution, which further renders between-patient comparisons possible. Finally, a B-spline classification of standardized distribution profiles using Ward's unsupervised method was performed. These grouped profiles were then compared to clinical observations.