3D Medical Image Registration using Mutual Information and Scatter Search

Abstract

Image Registration (IR) is a challenge that arises in many image processing applications when several images must be aligned. In particular, we treat the Medical Image Registration (MIR) case of different modalities. The big range of applications in medical imaging, goes from computer assisted diagnosis to computer aided therapy and surgery [2]. MIR is treated as an optimization problem with the goal of finding the spatial mapping that will bring a moving image into alignment with a fixed image. Deterministic algorithms are mainly used to solve it, together with some stochastic ones. A drawback of the latter is that many of them become stuck in local optima, especially in multimodal registration with several parameters [3], [4]. In this context, one of the most popular metric approaches is the Mutual Information based methods using deterministic optimization algorithms to compute the cost function with the mentioned problems. This work is aimed to overcome this disadvantage using the Scatter Search optimization algorithm [5] with the mutual information approach, proposed by Mattes et. al. [8]. The optimizer was tested and contrasted with RegularStep Descent Optimizer [9] and the 1+1 evolutionary algorithm [12]. Multimodal, rigid, 3D/3D, image registration of tomographic brain images was performed over a database available in RIRE 2 project. It was used the Insight Segmentation and Registration Toolkit (ITK) 3 , which is a set of libraries in C++ designed for the development of registration methods [10]. Qualitative and quantitative validation of the results are satisfactory, the results proves the accuracy and applicability of the proposal method comparing with conventional methods, and not being stuck in local optima.