Multi-modality Image Segmentation And Registration Of Pig Myocardial Infarction Model

With the development of medical imaging technologies in recent years, multi-modality image fusion has become a hot research field of medical imaging, which can provide more comprehensive pathology information to assist physicians in clinical diagnosis, developing reasonable treatment plans and assessing the effect of treatment. Ischemic heart disease, also known as myocardial infarction, is a disease with high mortality rate. The myocardium defect judgment of the patients with coronary artery disease has very important significance on the development of therapeutic schedule, predict efficacy and determine the prognosis.We use the myocardial metabolic imaging(PET/CT) and myocardial perfusion imaging(SPECT/CT) of pig myocardial infarction modal for image fusion, which can help the physician more accurately estimate the extent of myocardium defect. Because these two kinds of imaging both have a corresponding CT image, the research idea of this thesis is to develop method for heart CT image segmentation and registration, then combined them to achieve the fusion of CT, PET and SPECT images of the heart.Considering the characteristics of cardiac CT image, this thesis proposes an atlas-based segmentation and registration framework of heart and establishes a complete image fusion process. First we generate the atlas of heart by manually segment the CT image provided by the PET/CT scanner and thus obtain the corresponding PET image of the heart. Then we use the atlas-based segmentation and registration framework to segment the SPECT signal from the SPECT/CT image. After that we match the SPECT image of heart to the CT image by reverse registration. Finally we combine the three modalities of cardiac CT, PET and SPECT images to form a fused visualization.The core of the whole process of image fusion is the registration of cardiac CT images, the registration result directly determine the precision of the fusion result. First we use affine registration for coarse registration to preliminary align the hearts in two CT volumes. Then we use an improved Diffeomorphic Demons algorithm for non-rigid registration as the fine registration process. In this thesis, we also use multi-resolution registration strategies to improve the efficiency and accuracy of Diffeomorphic Demonsalgorithm.The proposed method is testified using data from the Fourth Military Medical University. The results show that the proposed atlas-based segmentation and registration framework of heart can segment the heart effectively and accurately. The improved Diffeomorphic Demons algorithm has higher registration precision. Fusion image shows the myocardial metabolic region and myocardial perfusion region of pig myocardial infarction modal, which can be used for observation and estimation of the extent of myocardium defect.