Supervised rigid image registration with CNN for MR brain images

Abstract

Image registration is an important and fundamental topic in medical image analysis applications. The objective of image registration is to get spatial transformation for the moving image to align with a reference image. In addition to conventional image registration methods, deep learning methods, which have gained popularity recently, give successful results by applying them to registration problems as well as many other computer vision problems. In this study, spatial transformation parameter estimation has been performed by utilizing a supervised deep learning method. The rigid method has been used as the spatial transformation model. A supervised convolutional neural network (CNN) estimates the rigid transformation parameters that are used for warping moving images to a fixed image. The most important contribution of this study is to perform supervised rigid image registration with the CNN regression model for MR brain images. In the study, it has been tried to determine the parameters that give the best results for the problem by trying different parameters.