—Registration or spatial normalization of diffusiontensor images plays an important role in many areas ofhuman brain white matter research, such as analysis ofFraction Anisotropy (FA) or whiter matter tracts. Moredifficult than registration of scalar images, spatialnormalization of tensor images requires two important parts:one is tensor interpolation, and the other is tensorreorientation. Current tensor reorientation strategypossessed many defects during tensor registration. Toovercome the shortcomings, we first presented amulti-channel model with one FA and six log-Euclideantensors, and then proposed an adaptive chaotic particleswarm optimization to find the global minima of theobjective function of the multi-channel model. The resultson 42 slices inter-subject registration indicate that ourproposed method can produce accurate and optimizedparameters of tensor registration with fastest speed relativeto Genetic Algorithm and Particle Swarm Optimization
展开▼
