While we have previously reported on multiscale segmentation of single-figure anatomic objects from medical images by deformable m-rep models, here we report on a method of segmentation of multiple geometrically related single-figure objects from a medical image. Inter-object geometric effects within segmentation of structures in the male pelvis from CT in the planning of radiation therapy of the prostate will be illustrated. Given a model in which each object is formed from a mesh of medial atoms, the method adds an object-complex similarity transformation stage ahead of the figural, medial atom, and boundary deformation stages of model transformation. The final three stages reward geometric interrelations among objects that occur in the model, according to the objectintrinsic coordinate systems provided by m-reps.
P. Thomas Fletcher, Stephen M. Pizer, A. Graham Ga