Comprehensive Analysis of Image Registration Techniques on Brain MR Images

Abstract

Medical image registration is an important preprocess of image-guided systems. Since image registration brings the images to the same coordinate system of the specified reference image, image registration should not be neglected to be able to make accurate comparisons between results obtained from medical images. Basically, registration is an optimization problem. The parameters of the specified transformation algorithm are optimized based on specified functions and parameters of registration. In this study, T1-weighted structural 3D brain MR images on IXI dataset have been registered into reference image by the affine transformation in the proposed registration method. During experiments, the effects of several parameters and functions on registration performance have been investigated with different preprocessing techniques applied to brain MR images. After several experiments, the most successful outcome of various experiments was achieved by using Powell optimization function along with Linear Interpolation, when applying Median Filter with CLAHE to images in the suggested registration method. The NCC was used to compare the registration results. The study's results demonstrate that the proposed registration method outperformed the widely-used registration tool SPM8 with mean NCC of -0.753. © 2023 IEEE.