The Research And Implementation Of Algorithm Of Medical Organization Image Segmentation

With the improvement of the imaging technology, medical images play a moresignificant role in medical diagnosis. Medical image processing becomes also naturallya current hot spot, in which, segmentation is an essential aspect. The imagesegmentation relates to the effect of such other aspects as diagnosis, registration,reconstruction and so on.Thanks to the complexity and particularity of the principle of the medical imaging,the traditional segmentation technologies no longer apply to the medical images withintensity inhomogeneity. The segmentation algorithms suitable for medical imagesbegin to be studied. For modern medical images, intelligence has also become theimportant factor in the medical image segmentation, in addition to the accuracy andstability. Based on the above, this paper will mainly research on the kind ofsegmentation algorithms, which can satisfy the features of the modern medical image.The major work is as follows:1. According to the characteristics of medical images, we sum up the commonand practical segmentation methods, which mainly include some traditional, classicalalgorithms and the practical algorithm that is often improved by scholars. The formerare simple and their applicability is not strong, but they are often used for reference byscholars. The latter can’t deal with medical images, but people try to improve orcombine them to put forward new algorithms.2. C-V model for Level set is introduced, which can’t apply to the MR imageswith intensity inhomogeneity. So some improvement is proposed: bias field model thatis estimated by some basic functions is introduced to C-V model, and the original pixelsare the product of bias field and true pixels. Based on the above ideas, the C-V modelbased on intensity inhomogeneity field is put forward, and the combination coefficientsfor bias field are calculated in the process of minimization of the energy function. Thealgorithm has the advantages of simpleness and fast convergence; what’s more, it can beused for the segmentation of MR images.3. A Level set segmentation algorithm based on intensity inhomogeneity in local region is research in detail. The character of Local region in MR images is analysed andis introduced to the algorithm in the third chapter. And bias field is estimated directlyrather than is obtained by fitting, which makes the result of estimation, correction andsegmentation more accurate. However, the iteration speed of the algorithm is slowerthan the algorithm of the third chapter.4. The principle and implementation process of the traditional MRF algorithm areintroduced in detail. It is not suitable for the segmentation of MR images. Therefore, thealgorithm in the fourth chapter is combined to propose a new MRF model, which is ableto use space statistics and gray information more fully to improve the accuracy of thealgorithm and application scope. Because the computational burden becomes heavier,some measures to decrease the iterations number of main process are taken to decreasethe computer time.