Diffusion tensor image registration based on demon-affine method and FSGA-BFGS algorithm

Registration or spatial normalization of diffusion tensor images plays an important role in many areas of human brain white matter research, such as whiter matter tract or analysis of Fraction Anisotropy (FA). More difficult than scalar images, spatial normalization of tensor images requires two important parts, one is tensor interpolation and the other is tensor reorientation. Current tensor registration strategy possesses many defects such as the low precision and low speed in non-rigid registration. To overcome those weaknesses, this paper firstly presents a Demon-Affine Registration method to improve the traditional DTI registration. Afterwards, in order to improve the efficiency and accuracy of search strategy, this paper proposes a new hybrid algorithm which merges the Broydon- Fletcher-Goldfarb-Shanno search into the Fitness-Scaling Genetic Algorithm framework as a basic operator. The Demon-Affine Registration method improves the model of Diffusion Tensor Image registration while the Fitness-Scaling Genetic Algorithm enhances the efficiency of the search procedures. The simulation results showed that the tensor registration results of our proposed method possesses higher accuracy and less time consumption.