Evaluation of a Statistical Shape Model for the Liver

A statistical shape model (SSM)for the liver has been proposed recently which is based on a cubic Hermite mesh with a small number of elements. By fitting to the data cloud of a liver surface, such a mesh is capable of capturing the complex liver shape with fine details. However, the liver SSM suffers from uncertain nodal correspondence in particular at locations where local salient shape features lead to incompatible/incorrect node correspondence. Furthermore, the fitting error is evaluated by the sum of all projection vectors without differentiating the fitting quality at local regions. In this work we assess the quality of node correspondence by using generalisation and specificity measures. Moreover we use a Jaccard index (JI)to evaluate the overall quality of fitting error. We found that 4-element mesh yielded the least fitting error, and the mesh also had a high generalisation.