Research On Medical Image Non-rigid Registration Algorithm

There are two type of image registration, rigid registration and non-rigidregistration. the rigid registration mainly resolve the simple image transformation (suchas translation, rotation, etc.) problem,the research of such transformation is relativelymature; non-rigid registration mainly solve the flexible image elastic transformationproblem,It allows the relationship between any two pixels could change during theregistration process. because of the difference of imaging mode, imaging time, andimaging individual, In actual clinical applications, the difference between the image isrelatively large. the rigid registration unsuitable to solve such problem. This paper studythe non-rigid registration algorithm about medical image, the work as follows:1. Demons algorithm which based on the optical flow field is a better non-rigidalgorithm, but in this algorithm, each pixel can move freely. if we directly depend on theoriginal image, at the time of the original image has relatively small difference in grayvalue, the different points may be mapped to the same location, this will cause imagetopology change. To solve this problem, this paper proposes a multi-resolution (three)strategy to curb pixel excessive offset, the specific process is as follows: low resolutionregistration transformation parameters sequentially is used as the initial registrationtransformation of the higher level registration. This method reduces the sensitivity ofthe pixel gray, ensures the stability of the image topology.2. Demons algorithm think registration process is mainly determined by two forces:the reference image gradient is the internal forces witch promote the image deformation;the gray difference between reference image and the floating image is the externalforces witch decide the direction of the deformation of the image. But only use gradientof the reference image as the internal forces to promote the image deformation is notenough, when we lack the gradation information we can not obtain the correspondingtransformation parameter, and only one internal force usually takes long time. In thispaper, based on the successful use of mutual information. we use both of the referenceimage gradient and mutual information as the internal forces to promote the registrationprocess.this improvement of image deformation algorithm, making registration accuracy and registration time improved a lot.3. By studying several non-rigid registration algorithm, comparing differentalgorithm, we found that the B-spline has local registration characteristics. in this paperwe use this characteristic, at first we use the improved Demons registration algorithmfor image coarse registration, then we use B-spline algorithms to obtained further fineimage. In this way we get a more accuracy image.