Julip Jung MS, Presenter: Nothing to Disclose

Helen Hong PhD, Abstract Co-Author: Nothing to Disclose

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

Lung tumor segmentation on chest CT images is essential and important to measure tumor size and volume for cancer diagnosis, treatment planning and therapy response assessment. One of the most challenging tasks is the separation of lung tumor from neighbor structures such as chest wall or mediastinum. Because lung tumor is often attached to chest wall or mediastinum and its intensity is similar to its neighbor structures. Thus, we propose a novel segmentation method of lung tumor on chest CT images using asymmetric multi-phase deformable model with peripheral region probability map.

Our method has been applied to ten patients with lung tumor. Each CT image had a matrix size of 512 x 512 pixels with in-plain resolutions ranging from 0.52 to 0.73 mm. The slice thickness ranged from 2.5 to 3.0 mm and the number of images per scan ranged from 103 to 158. First, optimal volume circumscribing lung tumor is decided by dragging inside the tumor and initial tumor region was extracted by applying thresholding with threshold value of -400HU to the optimal volume. Second, to estimate the possibility of neighbor structures to be separated from lung tumor, peripheral region probability map is generated. In peripheral region probability map, tumor region is suppressed by localizing lung tumor using Hough ellipse transform and neighbor structure region is highlighted by counting the number of voxels with intensity higher than 0HU. Finally, the initial tumor region is refined by asymmetric multi-phase deformable model with peripheral region probability map. For evaluation of our segmentation method, automatic segmentation results are visually assessed in axial plane and Dice similarity coefficient is measured between automatic and manually segmented tumor regions.

Our peripheral region probability map can estimate the possibilities of region to be separated from lung tumor. Our asymmetric multi-phase deformable model with peripheral region probability map can separate lung tumor from attached chest wall and mediastinum without lung segmentation.

Our segmentation method can be used in lung tumor measurements for cancer diagnosis, treatment planning and therapy response assessment.

http://abstract.rsna.org/uploads/2014/14017026/14017026_2tus.jpg

Jung, J, Hong, H, Goo, J, Automated Segmentation Method for Lung Tumor Attached to Neighbor Structures in Chest CT Images.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14017026.html