Non-rigid Registration of White Matter Tractography Using Coherent Point Drift Algorithm

Axonal fibers in the white matter are in charge of bio-signal delivery and relate information between neurons within the nervous system and between neurons and peripheral target tissues. Tract-based analysis (TBA) can directly bridge white matter and its connected cerebral cortex to achieve a joint analysis of the brain's structure and function. However, the accuracy of TBA is highly dependent on the quality of spatial registration of fiber bundles of different individuals to the standard space. In this paper, a non-rigid point registration, Coherent Point Drift (CPD), is applied for registration of fiber bundles. Both the fiber features and the registration accuracy are evaluated to determine the correspondence among fiber bundles. Experiment results on twelve real data showed higher registration accuracy of the proposed method on mean nearest neighbor distance and fractional anisotropy (FA) profiles than traditional registration methods, such as affine, elastic and Iterative Closest Point (ICP).