Sang-Joon Park MD, Presenter: Nothing to Disclose

Doohee Lee, Abstract Co-Author: Nothing to Disclose

Jin Mo Goo MD, PhD, Abstract Co-Author: Research Grant, Guerbet SA

Hyun-Ju Lee MD, PhD, Abstract Co-Author: Nothing to Disclose

Chang Hyun Lee MD, PhD, Abstract Co-Author: Nothing to Disclose

Chang Min Park MD, PhD, Abstract Co-Author: Nothing to Disclose

To develop an accurate segmentation technique for bronchial trees in the lung parenchyma and to investigate its performance compared with previously reported extraction algorithms.

Thirty-two COPD patients with GOLD stage 1 were selected for investigating the performance of the airway segmentation scheme in this study. The images were obtained with low-dose chest CT (40 mAs at 120 kVp) using soft reconstruction kernel (Sensation 16). A PC-based in-house software was developed for fully automated segmentation of the pulmonary airways using the following steps: 1) intensity adaptive region-growing (IARG) technique, 2) eigenvalues-ratio of the Hessian matrix, 3) vector-based feature identification (VFI) scheme with false positive (FP) reduction process. The performance of segmentation results from the proposed technique was evaluated with Hessian-based method and VFI without FP reduction scheme and was compared to that of IARG for all subjects.

The average branch count (BC) and tree length (TL) of 32 subjects were 104±28.6 (1679.3±500.7 mm) with only IARG method, 129.9±31.9 (1910.3±560 mm) with combination of IARG and Hessian methods, and 128.9±31.8 (1904.7±464.9 mm) with combination of IARG, Hessian method, VFI with FP reduction scheme, respectively. Segmentation results with combined methods extracted more branches compared with that of only IARG method (p=0.003, respectively), while the proposed and Hessian-based methods showed no significant difference each other in extracting BC (p=0.59). However, the percentage of FP voxels decreased by 57.5±16.6% after applying VFI with FP reduction scheme (p=0.001). Our system was equipped with an Intel i7 CPU at 3.4 GHz and 16 GB memory, and the mean time of the whole processing was 51.84±14.37 seconds.

By using novel airway segmentation techniques comprising VFI with FP reduction scheme, we could segment airway branches up to 128.9±31.8 BC and 1904.7±464.9 mm TL even if some cases showed difficult appearance for identifying small airways accompanying leakages. This study can be a vital role as a preprocessing step for regional analysis of pulmonary airways and their functions in the lung parenchyma for various lung diseases in the clinical environment.

This study could provide a method to segment the airways up to 58% FP reduction rates.

Park, S, Lee, D, Goo, J, Lee, H, Lee, C, Park, C, Segmentation of Bronchial Trees Using a Vector Stream Information in Low-Dose CT Images: Application to COPD Patients.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14010138.html