A completely automated processing pipeline for lung and lung lobe segmentation and its application to the LIDC-IDRI data base
2010
Automated segmentation of lung lobes in thoracic CT images has relevance for various diagnostic purposes like
localization of tumors within the lung or quantification of emphysema. Since emphysema is a known risk factor for lung
cancer, both purposes are even related to each other. The main steps of the segmentation pipeline described in this paper
are the lung detector and the lung segmentation based on a watershed algorithm, and the lung lobe segmentation based
on mesh model adaptation. The segmentation procedure was applied to data sets of the data base of the Image Database
Resource Initiative (IDRI) that currently contains over 500 thoracic CT scans with delineated lung nodule annotations.
We visually assessed the reliability of the single segmentation steps, with a success rate of 98% for the lung detection
and 90% for lung delineation. For about 20% of the cases we found the lobe segmentation not to be anatomically
plausible. A modeling confidence measure is introduced that gives a quantitative indication of the segmentation quality.
For a demonstration of the segmentation method we studied the correlation between emphysema score and malignancy
on a per-lobe basis.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
0
References
2
Citations
NaN
KQI