Evaluation on diffusion tensor image registration algorithms

With the application in many neuroimaging studies, diffusion tensor image (DTI) registration has generated considerable interest and been studied widely. Although a number of DTI registration methods have been developed, their performances have not yet been compared systematically. This work addresses this gap by comparing a large number of existing DTI registration methods and gives the comprehensive evaluation results. In this paper, the open-access IXI DTI dataset were used. In order to compare the accuracy of tensor matching, 11 open-source registration methods were evaluated with 7 quantitative and open-access evaluation criteria that measure the similarity among tensors (namely tensor-based techniques) or scalar images derived from diffusion tensors (namely scalar-based techniques). The evaluation results indicate that the diffeomorphic deformable tensor registration method (referred to as DTI-TK) is the best method, followed by the symmetric image normalization method (referred to as SyN).