Application Research On Medical Image Registration

With the rapid development of biomedical engineering and computer technology, medical imaging technology applies more and more extensively in clinical diagnosis and treatment, and many sorts of imaging equipment have been emerged, such as CT, PET, MRI, and US. Since different medical image can reveal different information of relevant organ or tissue, an increasing number of clinical activities require the comparison and analysis of images acquired using different modalities and scanners, from different subjects, or from the same subject at different times. Medical image registration is an actively researched image processing technique for recovering complex mismatches existing between such different types of images due to various reasons. Registration has been the hottest research point because of its significant clinical value. Medical image registration is often divided into two classes:features based registration and pixel intensity based registration. Intensity based registration method is now popular as it doesn’t require segmentation and feature extraction.This paper implemented3-dimentional rigid registration of medical images based on normalized mutual information, and improved this algorithm at terms of registration speed. For3D images, the amount of calculation is very great and this process is time-consuming. We address this issue by adopting a damping orthogonal algorithm and thus providing a proper initial value for the following optimization algorithm. The experimental results revealed that this algorithm performed successfully. We applied this registration method to change detection and characterization of brain tumor, and accomplished registration of MR image sequences.However, when analyzing images of soft tissues, such as heart, lung, etc, considering them as rigid body is unreasonable. To address this problem, this paper implemented hierarchical elastic registration of medical image. This elastic registration algorithm model the elastic transformation between images as an interpolation of multiple local rigid-body registrations. This approach involves subdivision of one or both participating images, followed by independent registration of corresponding subimages. Then we do many simulation experiments, and compare with rigid registration algorithm. This elastic registration algorithm is proved more precise.Real Time Myocardial Contrast Echocardiography (RT-MCE) is a new reliable technique in assessing myocardial microcirculation. So the quantitative analysis of RT-MCE has been one of the hottest research points in related areas. And the most popular analysis method is time-strength curve analysis. But the misalignment of region of interest results that this method is inaccurate. This paper applied elastic registration technique to RT-MCE quantitative analysis and realized the dynamic tracking of ROI.