The Segmentation Of Kidney Based On Digital Subtraction

In recent years, Medical equipment and systems which can be used for the detection, storage, transmission, analysis and display of images have a major development. To biologists, medical scientists and clinicians concerned, the application of these new technologies expand their biomedical research on the observation and measurement of photo imaging capabilities. Now the medical imaging software don't only apply high-quality image but also play a more important role in support of making a correct diagnosis. Urinary CAD software has been developed on the background of the market demand. It can be used in conjunction with multi-slice CT in hospital. So it can reduce doctors'intensity and raising the quality of diagnosis of renal function. Medical Image Segmentation is the base of subsequent operations such as 3D reconstruction, quantitative analysis of normal tissues and ill tissues. And it is also the crux of clinical application .The accuracy of segmentation is important for doctor to plans to make a correct diagnosis .For Urinary CAD, registration algorithm and subtraction algorithm are the key of much algorithm. It provides the feasibility to further analyze and diagnose kidney disease accurately, conveniently.In the paper, the digital subtraction angiography technology is innovatively used for the segmentation of kidney. After the segmentation of kidney from the fusion image of contrast image and plain image, doctors can analyze the morphology and function of the kidney. So it is of great significance for the diagnosis and treatment of kidney disease. Here is a simple understanding of digital subtraction angiography technique:(One) Digital Subtraction AngiographyDigital Subtraction Angiography (DSA)is a powerful technology which can get vascular visualization from X-ray series photo(chart one),it has been in clinical use for more than 20 years. DSA is based on the assumption that the organizations around the vein are location-unchanged and gray-unchanged. Since the early 1980s, the introduction of subtraction technique, clinic use has showed that the location-unchanged and gray-unchanged assumption were not feasible because the patient's movements can not be completely avoided. Then the"artifacts"appears, DSA is more difficult because of it. Some technologies which can improve the diagnostic value of subtraction photo by reducing artifacts have been in use. They include tomography,synthesis imaging technology and so on. But these technologies aren't in clinic use because they need materials either expensive or hard to produce. So we have to perish artifacts by image process and get correct subtraction. First, we can usually calculate the correspondence between the pixels in consecutive images. This is image registration which can be used to analysis images.(Two) Image RegistrationImage registration is a space transform relation between two images. Through this space transform, we can make an image (Floating image F) has a coherence on space location with the other image (Reference image R).The registration between image A and B is searching for a mapping relation T:X_A→X_B,which makes every point on X_A has unique corresponding point on X_B.For medical images, this correspondence is that the same anatomy point in human being have the same space location on two registration images. The registration result should make the anatomic points on the image, at least the points which have diagnostic signification matched. The images after registration could be on medical subtraction and segmented from different apparatus.Take image pixel gray as a random variable, we can use statistics and informatics to solve images registration. Mutual information which can describe the statistical correlation between the variables is a fundamental concept in information theory. In image registration, the mutual information is used to describe the statistical correlation between pixels in two images. We can adjust registration parameters through an iterative optimization algorithm. When the mutual information achieves maximum, the two images have a best space registration location. The mutual information is described as this formula: p(x,y) is joint probability distribution,p(x) and p(y) are the marginal probability distribution of two images.Later, Studholme proposed a final measure of an mutual information (Normalized Mutual Information):The maximum of normalized mutual information is searching for a transform to make joint entropy is the minimum comparative to margin entropy. On one side, it thinks over joint entropy is minimum; On the other side, it thinks over the information of overlapped images. At the same time, it has a good balance between the two. After measure function, we need a proper optimization algorithm to change parameters so that the comparability measure could achieve maximum. The problem of registration has been transformed multi-parameter optimization. The paper uses Powell algorithm as optimization algorithm:Powell algorithm is given by Powell in 1964.It came to being on researching of the minimum of the function:f(x)=1/2X~T QX+b~T X+c(Three) Mathematical morphologySet theory is used to describe mathematical morphology. Expanding and erosion are its foundation. We usually use the binary mathematical morphology. Every pixel in Binary image (black-and-white image) only has two grey{0,1}.The operation targeted of binary morphology is two set. Propose A is a set of images, B is structure element. Morphological operation is to use B of A for operation.Expanding: A⊕B={c∈E~n|c=a + b, a∈A, b∈B}Erosion: AΘB={c∈E~n|c + b∈A,(?) b∈B}Open: A(?)B= ( AΘB)⊕BClose: A·B= ( A⊕B)ΘBBecause of its complexity and medical image registration error, the subtraction image is always not perfect. Mainly to reflect: Kidney has small hollows inside; some burs occurs outside; background noise. In order to obtain better results in the renal division, we need to deal with morphology.The paper processes human being CT images Based on the above theory. First, we make registration of scan and strengthening 3D image. The measure function is normalized mutual information. Powell multi-parameter optimization algorithms and one-dimensional iterative search algorithm are used to estimate registration parameters. Then, we make subtract between the two images with the relative position. Use iterative approach to obtain threshold values. So, the kidney is segmented. Last, mathematic morphology is used as a post-processing. Test results show that the method has good kidney segmentation. At present, author's not read such literature in kidney segmentation. It is better than traditional regional growth and the division. After the separation of kidney by this method, doctors can use medical knowledge to be effective for detection of renal function. So, it can play an important supporting role. The algorithm has been successfully applied to the analysis of urinary CAD renal function.