Non-rigid registration between brain CT images and MRI brain atlas by combining grayscale information, point correspondence on the midsaggital plane and brain surface matching

Abstract

Abstract —This paper presents an image registration method to warp brain atlas of magnetic resonance image (MRI) onto computed tomography (CT) brain images. A new deformable model is proposed to add constraints on corresponding midsagittal planes (MSP). The point correspondences in the MSPs are built up from registered MSPs of the brain CT images and the atlas. The object function is a combination of mutual information, point correspondences between MSPs, and correspondence between brain boundaries. The parameters of deformable model are optimized using an iterative strategy: maximization of mutual information (MI) to find the deformable field, assigning point correspondence of MSPs to modify the deformable field, and Gaussian smoothing to make the deformable field continuous and smooth. To speed up, we adopt a multi-resolution scheme. Experiments show that the proposed method yields good results especially at the region around MSP. Keywords-component; Registration; Feature-constrain; Multimodal; Midsagittal plane.