When constructing an application system for bronchoscope navigation or lung nodule diagnosis, it is important to segment the tracheobronchial tree from CT while knowing the anatomical names of each bronchial branch Fig. 3.51, which are the terms used by clinicians.
Because the bronchi bifurcate in an almost fixed manner, it is possible to assign anatomical names using graph matching between the input tree structure and the graph structure of a bronchial bifurcation atlas (learning tree structure). Mori et al. (2000/2002)  reported on an anatomical labeling process based on graph matching. The matching is performed based on running direction information and constraint of parent branches. Labeling of bronchial branches has also been performed based on graph matching . An example of bronchi branching models is shown in Fig. 3.52.
There are many variations in branching patterns of bronchi. It is necessary to have branching pattern atlases and to develop an algorithm to use such atlases showing variations. Mori et al. (2005)  showed a method for selecting suitable atlases in anatomical labeling. This method divides the tracheobronchial tree into five parts:
Fig. 3.51 Atlas of bronchus 
(a) trachea, (b) right upper lobe area, (c) right middle and lower lobe area, (d) left upper lobe area, and (d) left lower lobe area. Branching pattern databases are created for each area. Graph matching is performed in each divided area and an atlas removal process is introduced for choosing the best atlas in each area. Figure 3.53 shows an example of such an atlas.
Another method is based on a machine learning approach . This method computes many features for each branch name and constructs the classifier that outputs anatomical names from input features.
Figure 3.54 shows an example of anatomical labeling.
Fig. 3.52 Examples of bronchus branching model
Fig. 3.53 Examples of bronchus branching model atlas. This figure shows atlases of the right middle and lower lobe area