Research On Fusion And Reconstruction Of Multimodal Medical Images

In the medicine science,different sensors,different imaging techniques or examination methods have their own advantages and disadvantages,and the information provided is also very different.In general,to get more image information from an image,a single imaging technique is not enough for a doctor’s diagnosis.In many cases,only medical images provide comprehensive and rich information to make a diagnosis of the patient’s condition.At present,an obvious trend of medical image development is to correlate multiple images in different modes by image registration and fusion technology,and obtain a single image with more information than that under a single mode.Make it contain more information about the internal organization.This thesis is based on the sequence image registration and fusion of CT and MRI image sequences which are widely used at present to obtain an image sequence which can obtain more information from multi-source images.Then the fusion image sequence is reconstructed in order to observe the tissues and organs of human body more intuitively and to observe the 3D structure of the fusion image sequence.This thesis focuses on the registration,fusion and 3D reconstruction of multi-modal medical images.Finally,a multi-modal medical image research platform is designed and integrated.The details are as follows:1.The preprocessing technology of medical image and the processing of CT and MRI image before registration;2.Two dimensional image registration algorithms based on feature and mutual information are studied,and the results of the two algorithms are compared.In this thesis,a normalized mutual information registration algorithm based on contour pyramid is proposed.The registration results between the image determination layers are obtained.Because the focus of this thesis is on the registration of sequence images,it is necessary to find the matching relationship between CT and MRI,so the algorithm of 3D sequence image matching is studied.The problem of matching between layers and coarse registration of sequence images is solved,and then the results of 3D registration are further optimized by using the normalized mutual information registration algorithm based on contour pyramid.Thus,an image registration algorithm based on 3D to 2D is proposed.3.On the basis of the registration results,the image fusion algorithm is studied,in which many kinds of image fusion algorithms based on multi-scale analysis are studied,and the experimental results are compared.An image fusion algorithm based on non-subsampled Contourlet transform(NSCT)is proposed.Finally,according to the characteristics of the brain image studied in this thesis,the NSCT algorithm is improved,and the principal component fusion algorithm based on NSCT is proposed.4.3D reconstruction of fused data,including surface rendering and volume rendering,is carried out.The 3D reconstruction technology based on GPU accelerated ray casting algorithm is studied,and the final reconstruction results are obtained.5.The above research results are integrated into the multi-modal medical image fusion and reconstruction platform,and the realization of several modules of the platform is explained simply,and finally the system test is completed.