Solitary Pulmonary Nodule Segmentation Based on the Rolling Ball Method

Background Accurate segmentation of nodules in the lung parenchyma has been an important goal in refining computer-aided diagnosis of lung disease, especially early lung cancer. Most established segmentation methods can effectively segment nodules, although nodules arising from/adjacent to the pleura remain a challenge, especially in patient populations where benign pleural disease, such as tuberculosis, is common. Accurate segmentation of solitary pulmonary nodules, which carry a defined risk of malignancy according to change in size and other factors, is of great importance. We have developed a segmentation method for areas of lung parenchyma containing solitary pleural nodules. Beginning with rough region growing segmentation of the lung parenchyma, we applied the split sample Davis-Putnam (DP) total least squares algorithm to optimize the profile. We used regression analysis to extract the best bowls of repair radius. Finally, we employed the rolling ball method to assess the outer contours of the lungs and show any irregularities, which would include nodules abutting or arising from the pleura. Our results show that this method is effective in detecting these nodules. The success rate of our automatic repair method is >80%, close to that of manual repair.