Shape based region growing using derivatives of 3D medical images: application to semiautomated detection of pulmonary nodules

This paper presents a new method for shape based segmentation of 3D medical images. 3D geometric information is calculated for each voxel by computing the partial derivatives of the 3D image. The shape features of the iso-intensity surfaces are subsequently extracted. The extracted shape features are combined with 3D intensity-based region growing to give accurate separation of connected objects having different shapes but similar intensity values. We have applied this method to both synthetic and real 3D CT lung images. The experimental results demonstrate that the presented method, unlike the traditional intensity-based method, is able to segment connected objects accurately. A sphere can be differentiated from a connected cylindrical shape within the synthetic data. In the case of the real 3D CT lung images, all of the nodules can be detected and separated accurately from adjoining blood vessel or from the lung wall.