Analysis of pediatric airway morphology using statistical shape modeling

Traditional studies of airway morphology typically focus on individual measurements or relatively simple lumped summary statistics. The purpose of this work was to use statistical shape modeling (SSM) to synthesize a skeleton model of the large bronchi of the pediatric airway tree and to test for overall airway shape differences between two populations. Airway tree anatomy was segmented from volumetric chest computed tomography of 20 control subjects and 20 subjects with cystic fibrosis (CF). Airway centerlines, particularly bifurcation points, provide landmarks for SSM. Multivariate linear and logistic regression was used to examine the relationships between airway shape variation, subject size, and disease state. Leave-one-out cross-validation was performed to test the ability to detect shape differences between control and CF groups. Simulation experiments, using tree shapes with known size and shape variations, were performed as a technical validation. Models were successfully created using SSM methods. Simulations demonstrated that the analysis process can detect shape differences between groups. In clinical data, CF status was discriminated with good accuracy (precision = 0.7, recall = 0.7) in leave-one-out cross-validation. Logistic regression modeling using all subjects showed a good fit (ROC AUC = 0.85) and revealed significant differences in SSM parameters between control and CF groups. The largest mode of shape variation was highly correlated with subject size (R = 0.95, p < 0.001). SSM methodology can be applied to identify shape differences in the airway between two populations. This method suggests that subtle shape differences exist between the CF airway and disease control.