3D non-rigid registration of volumetric data using thin plate spline

Abstract

Image registration is the process of determining the geometric transformation of a source image to match the target image. For a medical image registration, the source and target images are the cross-sectional images of human internal organs. The registration is required to compensate the motion of these organs. Since these organs are non-rigid by their nature, it is desirable to consider non-rigid transformation for registration. Also, although the images are 2D, the organs are 3D objects. Therefore, it is also desirable to consider 3D registration rather than a series of 2D registration. In this thesis, we investigate an effective and efficient 3D non-rigid registration for medical image data based on thin-plate spline method. The thin-plate spline is widely used mostly in 2D applications. We expand the framework into 3D and applied it to CT volumetric data sets of lung at two different phases: inspiration and exhalation. With a small number of control points within the 3D volume, the bending energy for matching is computed and applied to the entire volume within an interactive time. The results are promising that the proposed technique can aid physicians in many clinical applications.