A Robust B-Splines-Based Point Match Method for Non-Rigid Surface Registration

Small animal imaging provides a powerful tool to study cancer in animal models. To monitor therapeutic response, serial in vivo images are acquired at different time points. Accurate image registration is needed to improve the quantification of changes over time. However, mouse body is extremely flexible and deformable, mouse image registration is challenging. In this paper, we present a deformable image registration method for the whole mouse body. A non-rigid point matching method has been developed to align mouse bones which represent the posture of a mouse body. Deformation is modeled as a global affine transformation followed by a local B-splines deformation. The robust point matching method simultaneously estimates point correspondence and surface deformation. It does not need to identify and label feature correspondences. The method can handle complex three-dimensional surfaces with a large amount of points and has achieved a high computational efficiency. The method was tested on mouse microCT images acquired at different positions. The distance between the anatomical feature point pairs has decreased from 3.8plusmn1.0 mm for manually rigid-body registration to 0.9plusmn0.4 mm for non-rigid surface registration. We also demonstrated the surface matching method for tumor magnetic resonance images that were acquired in different phases of treatment. The non-rigid surface registration method works well for both bone and soft tissues. The registration method can be applied to not only small animal images but also human images in clinical applications.