Thoracic cavity segmentation algorithm using multiorgan extraction and surface fitting in volumetric CT

JangPyo Bae; Namkug Kim; Sang Min Lee; Joon Beom Seo; Hee Chan Kim

doi:10.1118/1.4866836

Thoracic cavity segmentation algorithm using multiorgan extraction and surface fitting in volumetric CT

Med Phys. 2014 Apr;41(4):041908. doi: 10.1118/1.4866836.

Authors

JangPyo Bae¹, Namkug Kim², Sang Min Lee², Joon Beom Seo², Hee Chan Kim³

Affiliations

¹ Interdisciplinary Program, Bioengineering Major, Graduate School, Seoul National University, Seoul 110-744, South Korea and Department of Radiology, University of Ulsan College of Medicine, 388-1 Pungnap2-dong, Songpa-gu, Seoul 138-736, South Korea.
² Department of Radiology, University of Ulsan College of Medicine, 388-1 Pungnap2-dong, Songpa-gu, Seoul 138-736, South Korea.
³ Department of Biomedical Engineering, College of Medicine and Institute of Medical and Biological Engineering, Medical Research Center, Seoul National University, Seoul 110-744, South Korea.

PMID: 24694139
DOI: 10.1118/1.4866836

Abstract

Purpose: To develop and validate a semiautomatic segmentation method for thoracic cavity volumetry and mediastinum fat quantification of patients with chronic obstructive pulmonary disease.

Methods: The thoracic cavity region was separated by segmenting multiorgans, namely, the rib, lung, heart, and diaphragm. To encompass various lung disease-induced variations, the inner thoracic wall and diaphragm were modeled by using a three-dimensional surface-fitting method. To improve the accuracy of the diaphragm surface model, the heart and its surrounding tissue were segmented by a two-stage level set method using a shape prior. To assess the accuracy of the proposed algorithm, the algorithm results of 50 patients were compared to the manual segmentation results of two experts with more than 5 years of experience (these manual results were confirmed by an expert thoracic radiologist). The proposed method was also compared to three state-of-the-art segmentation methods. The metrics used to evaluate segmentation accuracy were volumetric overlap ratio (VOR), false positive ratio on VOR (FPRV), false negative ratio on VOR (FNRV), average symmetric absolute surface distance (ASASD), average symmetric squared surface distance (ASSSD), and maximum symmetric surface distance (MSSD).

Results: In terms of thoracic cavity volumetry, the mean ± SD VOR, FPRV, and FNRV of the proposed method were (98.17 ± 0.84)%, (0.49 ± 0.23)%, and (1.34 ± 0.83)%, respectively. The ASASD, ASSSD, and MSSD for the thoracic wall were 0.28 ± 0.12, 1.28 ± 0.53, and 23.91 ± 7.64 mm, respectively. The ASASD, ASSSD, and MSSD for the diaphragm surface were 1.73 ± 0.91, 3.92 ± 1.68, and 27.80 ± 10.63 mm, respectively. The proposed method performed significantly better than the other three methods in terms of VOR, ASASD, and ASSSD.

Conclusions: The proposed semiautomatic thoracic cavity segmentation method, which extracts multiple organs (namely, the rib, thoracic wall, diaphragm, and heart), performed with high accuracy and may be useful for clinical purposes.

Publication types

Research Support, Non-U.S. Gov't

MeSH terms

Algorithms*
Cone-Beam Computed Tomography / methods*
Humans
Image Processing, Computer-Assisted / methods*
Imaging, Three-Dimensional
Mediastinum / diagnostic imaging
Pulmonary Disease, Chronic Obstructive / diagnostic imaging
Radiography, Thoracic / methods*