Multimodal Medical Image Registration, Fusion And Intelligent Diagnosis System Research

By using existent medical image processing, registration and fusion technologies, the registration method based on interpolation between slices, non-rigid registration method based on free form deformation, improved PCA fusion method and improved wavelet fusion method are proposed in the dissertation. On this basis, the (Myocardial Region Searching) MRS algorithm is developed to detect the heart region in SPECT myocardial perfusion images. In addition, an intelligent classification algorithm is presented for computer-aided whole body bone SPECT image intelligent diagnosis & report system, then the corresponding WBBSIID system is designed.Firstly, in the introduction part, the purpose and importance of our research are discussed, the registration & fusion technologies and their developments and the related achievements in the field in the world are reviewed, then the development direction and their applications are discussed.Secondly, as to the missed information between slices in CT and PET images, a cubic spline interpolation method is developed to restore the missed information, which is proved via experiments that it can improve the accuracy of the registration method. To solve the inconsistence problem existing in the CT and PET images, which is resulted from uncontrollable physiological heart beat and breath, free-form deformation (FFD) method based on an automatic recognition algorithm is developed to extract the feature points of PET and CT images, which can guarantee both the same deformation for two types of images and real-time running speed. In order to further improve the running speed, a novel method is first proposed to optimize the proposed registration algorithm using downhill searching method, which is more efficient than the traditional one.Thirdly, the evaluation criterions for the registration and fusion of medical images are introduced. An improved PCA fusion method is developed for enhancing the resolution of fused image. Additionally, an improved wavelet fusion method is also proposed to increase the wavelet coefficients proportion through weighted method, thus producing a high quality fused image with detailed information. The testing result via experiments using the proposed fusion methods demonstrates more effective than the traditional oneThen, a new MRS algorithm is developed for the recognition of heart region in...