Research Of Infant Brain MR Image Segmentation Framework Based On Image Registration Algorithm

Infant period is the human intelligence development and the critical period of brain development. Recent studies found the incidence rate, infant brain diseases showed a rising trend. Therefore, the correct diagnosis early exploration of infant brain diseases, has important significance in children with brain disorder rehabilitation. Due to the complex characteristics of infants with brain structure, usually require a doctor’s manual analysis of its. So, processing and analysis of automatic on infant brain image, quantitative analysis of brain structure is very important. The framework for automatic segmentation of infants brain MR image based on registration algorithm, is a prerequisite for accurate analysis of infant brain images, and provides valuable information for the diagnosis of doctors. Therefore, this thesis studies in the infant brain disease diagnosis, it has important scientific significance.This thesis studies the segmentation framework of infants brain MR image based on registration algorithm, the main research includes the infant brain image registration algorithm and establish the segmentation framework based on registration algorithm. This thesis analyses are as follows:(1) study of the traditional Demons algorithm and the active Demons algorithm. On this basis, combined with the orthogonal direction of gradient and curvature theory, the gradient direction of force and the orthogonal gradient force in the direction of synthesis of new forces, the active Demons algorithm is improved. (2) the infant brain image registration algorithm of two dimensional feature vector based on. Including multi-scale image structure, feature extraction, feature vector is constructed, matching feature points. The improved Surf algorithm based on 128 dimensional feature vector. (3) the Hammer registration algorithm for 3D infant brain images based on attribute vectors. A detailed study of the algorithm in the energy function, attribute vector, image deformation mechanism, smooth displacement field, multiresolution frame etc.. By the definition of driving voxel selection function and the use of thin plate spline interpolation, we improve the Hammer algorithm. (4) the segmentation framework of infants brain image based on registration algorithm. The basic principle is the template image registration and image registration is completed, to obtain a space transformation, and then use this space transformation to map the template image segmentation results to a sample image, so as to realize the automatic segmentation of the images of the samples.The experimental results show that, segmentation framework of infants brain MR image based on registration algorithm can realize the automatic segmentation of infant brain images based on the proposed. At the same time, prove the image registration algorithm is improved compared with the original algorithm, can achieve more accurate image registration.