Organ segmentation is a central topic in the field of medical image analysis, and a number of segmentation algorithms have been presented [142-145]. This section focuses on segmentation algorithms of organs/tissues in a human torso based on a CA model.
Since shape features of an organ play an important role in segmentation, researchers have attempted to incorporate them into their segmentation frameworks. Deformable model-based segmentation is a typical example. Pioneering works using a deformable model were reported in the early 1970s [146-148] followed by several epoch making works, Snakes , ASM , and active appearance model (AAM) , the first of which used a local shape feature, or curvature-based feature, to make extracted boundaries smooth and the latter two of which employed global shape or appearance features, or CA, to make extracted shapes more accurate anatomically. An alternative CA-based approach is atlas-based segmentation, which was initially developed for brain segmentation of magnetic resonance imaging (MRI) images and was then imported into organ segmentation of a human torso from CT, positron-emission tomography (PET), and MRI images.
This section describes CA-based segmentation algorithms for organs/tissues in a human torso starting with probabilistic atlas-based segmentation with an example of organ segmentation of a human trunk CT volume, followed by ASM, which employs
PDM, level-set-based segmentation with a shape prior, and ensemble learning-based segmentation with a shape prior.