Deformable Image Registration using Spring Mass System Jian-Kun Shen, Bogdan J. Mastuszewski, Lik-Kwan Shark Applied Digital Signal and Image Processing Research Centre, University of Central Lancashire, Preston, UK Christopher J. Moore North Western Medical Physics, Christie Hospital NHS Trust, Manchester, UK Abstract The paper describes a novel multi-resolution registration method. It is fast, robust and offers high registration accuracy. The algorithm models defor- mations using an elastic spring mass system, which contains sparse masses interconnected by springs. The proposed method uses data intensity values to guide deformation with local constraints imposed by interaction of inter- connecting springs. Moreover, by using such system prior information about the data can by easily embedded into the system to improve the registration accuracy. The performance of the method is tested using simulated as well as real dynamic magnetic resonance image dMRI data. 1 Introduction Image registration is one of fundamental tasks in image processing. This task can be simply considered as a process of aligning/matching two or more images having similar contents in some sense. For example, the images could have been captured at different times, from different viewpoints and/or using different types of sensors. Image registra- tion has been researched extensively in the last twenty years. Whilst global rigid/affine registration has matured, deformable registration is still under intense research, specially in the biomedical image processing [5] [6]. The image registration methods can be broadly divided into two main categories, feature-based and intensity-based registration methods. The feature-based registration methods requires pre-processing step to extract corresponding image features such as points, lines and curves. By matching the corresponding image features, deformation of the whole image can be calculated using one of “smooth” interpolation methods. The often used interpolation methods are B-spline [7], Thin-Plate Spline [3], Radial Basis Functions [1], Inverse Distance Weighted Interpolation [15] or their combination [8]. The intensity-based methods operate directly on image intensity values. One of the most pop- ular methods is to calculate the transformation using a set of equally spaced sparse control points, which are not linked to any specific image features, and finding the extreme in the 1 BMVC 2006 doi:10.5244/C.20.122