Brain Tissue Data Registration based on Context Demons Method

Image registration of brain tissue data has been used in disease diagnosis, surgical navigation, human brain mapping and some other fields for decades. At the present stage, brain image registration process generally adopts nonlinear registration algorithm, among which Demons algorithm is one of the most widely used. Owing to the growing processing requirement of large amount three-dimensional brain tissue data, the calculation burden of registration algorithm should be relieved. In this paper, the distribution of brain image intensities would be limited to a few significant categories by the mean clustering segmentation, which can facilitate the registration speed while ensuring the image registration effect. Then, an adjacent slice context processing configuration is proposed to reduce the redundancy calculation based on the slow spatial variation of brain tissue, and significantly improvement of the convergence rate in the Demons registration can be achieved. Finally, in view of the large tissue differences, an elimination and re-registration procedure is implemented for better accuracy, and a well preparation for the subsequent individual differences (such as lesion, tumor etc.) analysis can be achieved.