| Popis: |
Background: Spatially associated airway remodeling assessed by CT in COPD and asthma is of increasing interest as a disease sub-phenotype. Anatomical labeling in human airway trees is well defined up to the segmental level, but assignments at the sub-segmental leveIs are less well defined and manual labeling is time intensive and error prone. Aims: 1) Establish that sub-segmental branches follow a common spatial orientation pattern across the population. 2) Develop an automated algorithm for the assignment of sub-segmental labels. Methods: The average sub-segmental branch orientation is found by building a population average over a training database of 300 insp. CT scans (SPIROMICS study). Labels are applied to a subject by comparing against the population average and finding the spatially closest labels. Results: Sub-segmental airways exhibit a common spatial orientation. The median deviations in 300 subjects along 6 major paths are: LB1 ±30.5°, LB5 ±30.1°, LB10 ±25.2°, RB1 ±19.0°, RB5 ±28.2°, RB10 ±31.3°. Hence a reliable automated labeling is possible. Two human experts reviewed 30 airway trees and judged if the algorithm chose the correct branches for labeling. 95.2% of the selected branches were found to be correct. The two readers agreed on the classification of 561 branches and disagreed with each other on 12 branches (Fleiss9 Kappa=0.69). The wrongly selected branches were evenly distributed across the lung. Branches with segmentation errors were the main reason for wrong selections. Conclusions: We have developed a reliable method to include sub-segmental branches in an automated labeling system of the human airway tree. This will provide the opportunity to study the spatial dependence of airway phenotypes. |
