Fig. 12

Statistical shape models. To restrict the segmentation to a set of admissible liver shapes, a shape database is compiled, from which any new liver shape is expressed by a set of parameters called modes of variation. The various modes of variation (roughly 30 modes) are adjusted to fit the liver shape on image features. Statistical shape models impose hard restriction on the segmentation outcome by integrating prior shape. However, training data cannot capture all variations and therefore are sometimes too limiting to accurately model specific livers [13]