Postmortem Liver SSM and Liver Segmentation
Synthesis-based learning was successfully applied to train a postmortem liver SSM in Sect.188.8.131.52. The advantage of the proposed algorithm lies in the simple simulation of the shape transformation from an in vivo liver to a postmortem one. However, the approach may not simulate all possible shape changes, because of absence of physical point of view. Since the performance of the SSM strongly depends on the quality of synthesized liver labels, physics-based deformation approaches should be applied to synthesize liver labels in the near future, such as the finite element method (FEM)-based transformations . An interesting future topic would be comparison with an FEM-based approach and integration the constructed SSM with a postmortem liver segmentation algorithm.
The current liver segmentation performance for a cadaver is inferior to that for an in vivo patient [177-179,188] owing to postmortem-specific shape deformation and larger location variation. Extension of the EM algorithm before MAP segmentation to estimate shape and location will be an important endeavor.