A CI feature-based pulmonary nodule segmentation using three-domain mean shift clustering

A novel and more effective algorithm used for segmenting pulmonary nodules in thoracic spiral CT images was presented. The algorithm is based on mean shift clustering method and CI (Convergence Index) features, which can represent the multiple Gaussian model of pulmonary nodules both for solid and sub-solid, substantially. The algorithm has the following steps: (1) calculating the CI features of all pixels in the region of interest (ROI), (2) combining the CI features with the intensity range and the spatial position of the pixels to form a feature vector set, (3) grouping the feature vector set to clusters with mean shift clustering algorithm. Owing to our algorithm can represent the multiple Gaussian model both for solid and sub-solid nodules, it can be used in any user interested nodule regions, especially suitable for the segmentation of sub-solid nodules. Experiments demonstrated that our algorithm can figure out the outline of pulmonary nodules of different forms more precisely.