Segmentation Lung Fields in Thoracic CT Scans using Manifold Method

When the pathologies are in the close vicinity of the lung wall, the acquisition of the pulmonary nodules depends on accurate segmentation of the lung fields. However, the traditional methods based on pixels intensity cannot segment out them correctly. The paper proposes an effective segmentation method based on primary component analysis (PCA) manifold. It used the lung fields’ relationship in a lung to construct the shape manifold with B-spline interpolation. In the manifold space, according to the position of the affected lung field, a measurement had been used to find an amended position, and it was projected back into the shape space to reconstruct the prior shape. The shape was registered with the affected one and then segmented the original lung section to obtain the correct lung field. The experiment results illustrate that the proposed method has more correct segmentation ability than the methods based on rolling-ball and pixel intensity.