Global structure-preserving voxel classification for patient-specific surgical simulation

Summary form only given. Voxel Classification is a mapping of feature vectors to a discrete set of tissue classes. Surgical simulation is a technique whereby a user interacts, through virtual surgical tools, with an anatomical model endowed with realistic constitutive properties. Up to now, such virtual anatomical models have been obtained by elaborate manual segmentation. This paper suggests a minimally supervised procedure for extracting from CT and MR scans, co-registered in a distortion-tolerant manner, a highly descriptive classification of tissues which exploits global contiguity and prior assumptions about spatial extent and embedded structure, in order to generate a patient-specific anatomical model.