Conference proceeding
Graph-based 4D lung segmentation in CT images with expert-guided computer-aided refinement
2013 IEEE 10th International Symposium on Biomedical Imaging, pp.1312-1315
04/2013
DOI: 10.1109/ISBI.2013.6556773
Abstract
A new graph-based approach for simultaneous segmentation of lungs in 4D CT scans is presented. The approach is based on a "just enough" user interaction principle and consists of two stages. First, a fully automated graph-based segmentation algorithm is applied. Second, the user inspects the result and can correct local segmentation errors with all interactions performed within the graph-based computer-aided computational framework. The method was evaluated on ten 4D CT scans of lungs with disease (cancer, etc.). Compared against an independent reference standard, the average Dice coefficient was 0.966 ± 0.014 and 0.974 ± 0.009 after automated segmentation and subsequent interactive computer-aided refinement, respectively. Overall, fifteen out of twenty left/right lungs at inspiration and expiration needed refinement. This process took 4.3 min on average. The achieved improvement in segmentation performance was found to be significant (p <;<; 0.001). Results demonstrate good performance of the fully automated segmentation approach, which can be further improved by means of graph-based refinement.
Details
- Title: Subtitle
- Graph-based 4D lung segmentation in CT images with expert-guided computer-aided refinement
- Creators
- Shanhui Sun - University of IowaMilan Sonka - University of IowaReinhard R Beichel - University of Iowa
- Resource Type
- Conference proceeding
- Publication Details
- 2013 IEEE 10th International Symposium on Biomedical Imaging, pp.1312-1315
- DOI
- 10.1109/ISBI.2013.6556773
- ISSN
- 1945-7928
- eISSN
- 1945-8452
- Publisher
- IEEE
- Language
- English
- Date published
- 04/2013
- Academic Unit
- Roy J. Carver Department of Biomedical Engineering; Electrical and Computer Engineering; Radiation Oncology; Fraternal Order of Eagles Diabetes Research Center; Injury Prevention Research Center; Ophthalmology and Visual Sciences
- Record Identifier
- 9984186690502771
Metrics
12 Record Views