Hyun Hee Jo, Presenter: Nothing to Disclose

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

Jin Mo Goo MD, PhD, Abstract Co-Author: Research Consultant, Infinitt Healthcare Co., Ltd

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

We have developed an automatic segmentation method of pulmonary nodules in CT images using intensity, curvature and morphology information for the diagnosis of malignancy and therapy monitoring.

To measure a growth rate of pulmonary nodules in CT images for the diagnosis of malignancy and therapy monitoring, nodule segmentation is an essential preprocessing step. However, automatic segmentation is difficult due to density similarity of nodules and their neighbor structures as well as their various morphologies. In this paper, we propose an automatic nodule segmentation method in CT images using intensity, curvature and morphology information.

The CT images of forty patients with 44 pulmonary nodules (mean diameter 13.2±5.9mm) were acquired. An initial nodule segmentation using 3D seeded region growing was performed to extract a nodule candidate which includes the nodule and its neighbor structures. A nodule type (isolated, juxtapleural, juxtavascular or pleural tail nodules) was determined by calculating a boundary coverage ratio of the nodule candidate. The nodule candidate was refined by eliminating neighbor structures such as chest wall and vessels. For the juxtapleural nodule, connections between the nodule and chest wall were separated by linking maximum curvature points of the nodule candidate boundary. For the juxtavascular nodule, connections between the nodule and vessels were separated by using binary erosion with circular structuring element of minimum radius of the nodule candidate. For the pleural tail nodule, the segmentation algorithms of juxtapleural and juxtavascular nodules were successively performed. To evaluate the performance of proposed method, Dice coefficient(DC) between automatically(A) and manually segmented volumes by two radiologists(B, C) was measured. The DC was 0.95±0.06 for A vs. B, 0.92±0.06 for A vs. C, and 0.95±0.03 for B vs. C, respectively.

Our segmentation method can extract various pulmonary nodules without leakage to the neighbor structures by automatically dividing nodules into four types and applying different algorithms using intensity, curvature and morphology information.

Jo, H, Hong, H, Goo, J, Park, C, Automatic Pulmonary Nodule Segmentation in CT Images Using Intensity, Curvature and Morphology Information.  Radiological Society of North America 2010 Scientific Assembly and Annual Meeting, November 28 - December 3, 2010 ,Chicago IL. http://archive.rsna.org/2010/9014939.html