Medical Image Registration Using Thin-Plate Spline for Automatically Detecting and Matching of Point Sets

This paper presents a non-rigid registration method for medical images by applying thin-plate spline (TPS) transform. The thin-plate spline which has special spline properties has been widely used in non-rigid image registration. TPS transform registration scheme based on landmarks often extracts two sets of landmarks from different images. Corners often represent critical information to describe object features, and they are usually used as landmarks in image registration. The corner detecting method proposed in this paper is performed according to local properties (such as local maxima curvature) and global properties (such as the relationship between neighboring features in the contours). This method can also distinguish obtuse corners from round corners, and obtuse corners are much more valuable and useful for representing the shape of objects. The feature points are matched with coherent point drift method based on probability density estimation. This method can be used in both rigid and non-rigid point set registration. The proposed method is applied in CT and CBCT image registration, and achieves ideal results.