Brought to you by
ECR 2013 / C-2474
An automatic system for segmentation, matching, anatomical labeling and measurement of airways from CT images
Congress: ECR 2013
Poster No.: C-2474
Type: Scientific Exhibit
Keywords: Chronic obstructive airways disease, Computer Applications-3D, CT, Respiratory system
Authors: J. Petersen1, A. Feragen1, M. Owen2, P. Lo3, M. M. W. Wille4, L. H. Thomsen5, A. Dirksen5, M. De Bruijne6; 1Copenhagen/DK, 2Waterloo, ON/CA, 3Los Angeles, CA/US, 4Hellerup/DK, 5Gentofte/DK, 6Rotterdam/NL


The segmentation method has been used on 9711 low dose CT images from the Danish Lung Cancer Screening Trial (DLCST). Manual inspection of thumbnail images revealed gross errors in a total of 44 images. 29 were missing branches at the lobar level and only 15 had obvious false positives. A thorough inspection of 10 randomly selected images, revealed the method extracted 174 branches on average and only 3.79% of the found centreline (excluding trachea and main bronchi) to be partially incorrect (Lo et al (2009)).


The extracted wall surfaces were compared to manual annotations in 319 reformatted slices extracted perpendicular to the centreline at random positions in 7 subjects. Results show an average Dice's coefficient of 89%. The COPD gene phantom was scanned with the DLCST protocol and our segmentation method estimated all interior and exterior diameters within 0.3 mm of their actual values. Fig. 5 shows a segmentation result. Lumen volume and wall area percentage measured in a random sub-population of DLCST (480 subjects at two time points) using the method was reproducible and significantly correlated with lung function (Petersen et al. (2011)).


Airway branch matching was initially investigated in Petersen, Gorbunova et al. on data from a randomly extracted sub-population of the DLCST consisting of 237 subjects scanned 5 times. Limiting measurements to branches matched in all scans of the same subject was seen to significantly increase their reproducibility, see fig. 6.


The anatomical branch labeling tool was validated (Feragen et al. (2012)) on a subset of 20 subjects, 5 of each category: asymptomatic, mild, moderate and severe COPD. The average inter-expert agreement of two trained observers in placing labels L1-L10 and R1-10 was found to be 71%, whereas the system reached 72.7%. Reproducibility of the experts in repeat scans of the same subject, assessed using image registration, was 72.6%, the system reached 76%. Accuracy did not become significantly lower in subjects with severe disease. Fig. 7 shows an airway
tree with automatically assigned labels and a developed tool in which the labels can be manually corrected if needed.


The framework has been used to investigate the effect of inspiration on the airway dimensions of the DLCST subjects with normal lung function. Results will be separately presented at ECR. Moreover work is ongoing to use the framework to investigate the association between airway distensibility and COPD.

POSTER ACTIONS Add bookmark Contact presenter Send to a friend Download pdf
2 clicks for more privacy: On the first click the button will be activated and you can then share the poster with a second click.

This website uses cookies. Learn more