Medical Image Registration And Fusion Research

Medical image registration and fusion technique is a research hotspot of medical image processing these days. By registering and fusing the intra-modality and inter-modality medical images, doctors could diagnose and surveil the progress of illness and set down the more reasonable treatment plan.In this paper two rigid registration methods were described. One was intra-modality medical image registration based on projection, the other was inter-modality medical image registration based on mutual information. Projection algorithm is presented as follows: first, characterize the reference image and the float image and turn into binary images; then, project the binary images onto the abscissa and y-axis and get the projection vectors; last, according to the elements of the vectors, the rotation and translation parameters were calculated separately. The method was proved to be a fast and validate registration algorithm by FMRI images. Mutual information algorithm use mutual information as a new matching criterion to measure the statistical dependence between the image intensities of corresponding pixels in both images, which is assumed to be maximal if the images are geometrically aligned. The method does not require any prior assumption, medical knowledge, segmentation, feature extraction or other preprocessing steps. Powell algorithm which does not compute gradients and Brent algorithm were used to search the optimal rotation and translation parameters, while multi-resolution searching strategy leads to a fast image registration. The registration results of computed tomography(CT) and magnetic resonance(MR) are validated to be subpixel based on mutual information.The fusion of registering CT and MR images by data level fusion or feature level make the information complementarity each other, which realized the anticipative intention.