Computer assisted detection of abnormal airway variation in CT scans related to paediatric tuberculosis.

Abstract

Airway deformation and stenosis can be key signs of pathology such as lymphadenopathy. This study presents a local airway point distribution model (LA-PDM) to automatically analyse regions of the airway tree in CT scans and identify abnormal airway deformation. In our method, the airway tree is segmented and the centreline identified from each chest CT scan. Thin-plate splines, along with a local mesh alignment method for tubular meshes, are used to register the airways and develop point distribution models (PDM). Each PDM is then used to analyse and classify local regions of the airway. This LA-PDM method was developed using 89 training cases and evaluated on a 90 CT test set, where each set includes paediatric tuberculosis (TB) cases (with airway involvement) and non-TB cases (without airway involvement). The LA-PDM was able to accurately distinguish cases with airway involvement with an AUC of the ROC classification (and 95% confidence interval) of 0.87 (0.77-0.94) for the Trachea-LMB-RMB region and 0.81 (0.68-0.90) for the RMB-RUL-BI region - outperforming a comparison method based on airway cross-sectional features. This has the potential to assist and improve airway analysis from CT scans by detecting involved airways and visualising affected airway regions.