Study Of Landmark-Based Affine Registration Of Human Heart From Diffusion Tensor Imaging

In recent years, statistics show that cardiovascular disease has become the most dangerous disease for eight years. Heart maintains the momentum of the circulation of blood and many cardiovascular diseases, such as myocardial infarction and heart failure, is closely related to the structure of myocardial fibers. The present study of myocardial fibers have a lot of ways, but the diffusion tensor magnetic resonance imaging (DT-MRI) is the only way that can images vivo myocardial three-dimensional fiber structure noninvasively. DT-MRI is mostly used to study the cerebral white matter fiber structure, and DTI is gradually applied to the research on myocardial fibers structure in recent years.Different mode, resolution, individuals of the heart or the single heart beat will make heart images appear bigger difference, and this will affects doctor’s diagnosis of cardiovascular disease. Therefore, we need to register the different heart diffusion tensor images to make its space position alignment, so that doctor will be able to compare it with heart atlas or other hearts. Moreover, we can register multimodal images and then fuse these images to get more information and improve the efficiency of diagnosis. In addition, we can build an average model of human myocardial fiber structure through registration of different individual heart from diffusion tensor imaging and it will provide more reliable data about the pathological study of cardiovascular disease.Currently available image processing algorithms of diffusion tensor images are mostly for brain imaging, but these algorithms cannot be directly used in cardiac imaging. Diffusion tensor image registration based on the heart is more complex, different brain DT-MRI data have the similar size, shape and direction, but different myocardial fibers, especially for the ex-vivo myocardial fibers, have a big difference on size, shape and orientation of myocardial fiber, which can easily make the 3D heart image affine registration dropping into local optimal solution and get a bad result. Therefore, according to the characteristics of the cardiac diffusion tensor images, we propose the landmark-based affine registration algorithm for the heart from diffusion tensor imaging. According to the geometric characteristics of landmarks, the algorithm can adjust the size and orientation of the heart quickly, and the algorithm apply the mutual information as the similarity metric, use 3D affine transformation as transformation mode, then optimized by Powell optimization algorithm. Finally, we will reorient diffusion tensor to preserve the tensor orientation consistent with the surrounding myocardial fibers structure of the image. The experimental results show that the algorithm overcome the problem that optimization is easy to dropping into local optimal solution. The accuracy of registration result has reached the requirement of rigid registration and the result can provide an ideal initial position for non-registration. In addition, the adjustment of heart image size scaling method is put forward. We set two scaling factor, instead of single scaling factor, on transverse and vertical axis direction to zoom in and the scaling method can solve the problem that the original scaling method can’t get registration due to the big difference between transverse and vertical axis size.