Program Information
Automated Lung Segmentation Method Using Atlas-Based Sparse Shape Composition with a Shape Constrained Deformable Model
J Zhou1*, z yan2 , s zhang3 , B Zhang4 , G Lasio5 , K Prado6 , W D'Souza7 , (1) University of Maryland School of Medicine, Millburn, NJ, (2) Rutgers, the State University of New Jersey, Piscataway, NJ, (3) University of North Carolina at Charlotte, Charlotte, NC, (4) University of Maryland School of Medicine, Baltimore, MD, (5) University of Maryland School of Medicine, Baltimore, MD, (6) University of Maryland School of Medicine, Baltimore, MD, (7) University of Maryland School of Medicine, Baltimore, MD
Presentations
SU-F-BRF-2 Sunday 4:00PM - 6:00PM Room: Ballroom FPurpose:
To develop an automated lung segmentation method, which combines the atlas-based sparse shape composition with a shape constrained deformable model in thoracic CT for patients with compromised lung volumes.
Methods:
Ten thoracic computed tomography scans for patients with large lung tumors were collected and reference lung ROIs in each scan was manually segmented to assess the performance of the method. We propose an automated and robust framework for lung tissue segmentation by using single statistical atlas registration to initialize a robust deformable model in order to perform fine segmentation that includes compromised lung tissue. First, a statistical image atlas with sparse shape composition is constructed and employed to obtain an approximate estimation of lung volume. Next, a robust deformable model with shape prior is initialized from this estimation. Energy terms from ROI edge potential and interior ROI region based potential as well as the initial ROI are combined in this model for accurate and robust segmentation.
Results:
The proposed segmentation method is applied to segment right lung on three CT scans.
The quantitative results of our segmentation method achieved mean dice score of (0.92-0.95), mean accuracy of (0.97,0.98), and mean relative error of (0.10,0.16) with 95% CI. The quantitative results of previously published RASM segmentation method achieved mean dice score of (0.74,0.96), mean accuracy of (0.66,0.98), and mean relative error of (0.04, 0.38) with 95% CI. The qualitative and quantitative comparisons show that our proposed method can achieve better segmentation accuracy with less variance compared with a robust active shape model method.
Conclusion:
The atlas-based segmentation approach achieved relatively high accuracy with less variance compared to RASM in the sample dataset and the proposed method will be useful in image analysis applications for lung nodule or lung cancer diagnosis and radiotherapy assessment in thoracic computed tomography.
Contact Email: