Segmentation, anatomical labeling, branchpoint matching, and quantitative analysis of human airway trees in volumetric CT images

Solutions to four critical steps on the way to a fully automated evaluation of human lung CT scans are presented: (1) segmentation of human airway tree, (2) anatomical labeling of human airway tree, (3) matching of branch-points in the intra- and inter-subject cases, (4) quantitative measurements of cross-sectional planes along airway-segments.;Airway-segmentation fully works in 3D, incorporates knowledge of the shape of airways, and consequently offers a solution to the longstanding problem of leaking into the surrounding lung parenchyma. Low-dose scans and scans from diseased lungs can be segmented---tasks that were previously difficult or impossible to accomplish. The user does not have to tune any parameters. Segmentation time amounts to 3 to 10 minutes.;The anatomical labeling algorithm fully automatically assigns all 33 commonly used anatomical names to their respective airway-segments. The algorithm works on in-vivo trees. No user interaction is required, and the computing time amounts to about five seconds. An average of 97.1% of the assigned segment-labels prove to be correct (average number of assigned labels per tree: 27).;The branch-point matching algorithm finds corresponding branch-points between trees of two scans from the same patient (inter-subject matching is performed via anatomical labeling). The algorithm performs well on in vivo data (a comparable algorithm that works on in-vivo data was never presented before). No user interaction is required. Computing time amounts to two seconds for two full-sized trees. Validation showed an accuracy of 92.9% (average number of matches per tree-pair: 22).;Fully automated quantitative measurements are taken on airway cross-sections at any position along the airway-segments, replacing labor-intensive and error-prone manual delineation. Minor and major diameter and cross-sectional area are measured. The method works with sub-voxel accuracy for all airway orientations.;The complete system is built modularly. A framework based on XML files is presented that allows standardized data-exchange between individual modules. From these XML files all tree-related data can easily be retrieved for further evaluation like for example for physiological studies.