In this subsection, the outline of one of the most fundamental methods for the segmentation of the organs in medical images, active shape model (ASM) [1], is described along with other techniques.

Outline of ASM

The ASM segments a target organ region in a given image by registering a statistical shape model (SSM) of the boundary of the target organ to the given image. A SSM parametrically represents the statistical variety of the shapes of the boundary of the target organ, and the model is registered to the given image by estimating the values of the parameters of the SSM. One can divide the ASM algorithm into two steps: (1) construction of a SSM and (2) registration of the SSM. The basics of outlining the ASM are similar to those of many other methods for medical image segmentation. Figure 2.1 shows the process of (1) construction of a SSM and of (2) of registration.

Fig. 2.1 Flow of the construction of a SSM and of its registration to given images

1. Cootes and Taylor [2] first employed a point distribution model (PDM) to represent the boundary of a target. The boundary of the target organ in each set of training images is manually labeled, and a set of corresponding points is distributed on each of the labeled boundaries. The SSM of the PDM represents the probability distribution of the coordinates of the points. The statistical model has shape parameters that are the statistical random variables, and one can vary the shape of the model by changing the values of these parameters.

2. Registration of SSM

The objective of ASM is to determine the boundary of a target region in a new given image by registering the constructed SSM to the image. The SSM is registered by estimating appropriate values of the parameters of the SSM. When the SSM is registered accurately, each point on the surface represented by the SSM is located on the voxels around which the local image pattern has some features specific to the boundary of the target organ.