The Research And Application Of Brain MR Image Segmentation Based On Probability Density Weighted Geodesic Distance

The rapid development of medical imaging technology has made it become a major aid for clinical medicine.With the high resolution of soft tissue,easy to identify and quantitative analysis the structure of the brain,the magnetic resonance imaging(MRI)become the.key method of medical image segmentation,it widely used in research of brain lesions and neurotomy.Therefore,brain MR image segmentation has aroused a great deal of attention,and become a challenge in the research area.However,medical images are vulnerable to external factors such as noise and magnetic fields,these will lead to obscure and asymmetrical of images.Besides,comparing to natural images,the composition of human tissue is more complex,it requires more accurate thresholds for segmentation.All of above bring aboutobstacles to the development of medical image segmentation research.Superpixel is a fundamental yet vital and promising technique for many medical image processing tasks,e.g.brain MR image segmentation,due to the fact that it can significantly decrease the workload of post-processing steps.However,most of the existing superpixel algorithms cannot achieve accurate segmentation for the weak-boundary case where the intensity distribution intervals of different brain tissues have an overlap.This article starts from the characteristics of brain MR images,makes full use of the prior art knowledge of brain MR images.Observing the general structure of brain tissues in a given brain MR image,we overcome the difficulty by guessing the probability of each pixel in which it belongs to one of the categories,rather than depend on only intensity hints.Then a probability density based weighted geodesic distance is proposed in this article.The fuzzy c-means method is applied as subsequent processing to obtain the classification of the brain image.The algorithm creatively integrates the probability density function and designs a more efficient gradient calculation method.This makes the contrast between different tissues of the brain more obvious and each small boundary can be distinguished.The main work and innovation:1.Histogram statistics-based preprocessing and use the obtained sample values as a priori to estimate the probability density.Because the intensity distribution intervals of different brain tissues have an overlap.It is difficult to detect the exact division boundaries of different brain tissues,resulting in a fuzzy degree for class each pixel belongs to,and hard to directly segment.Therefore,in this paper,we first perform histogram statistics and then uses the obtained sample values as prior knowledge to estimate the probability density of each pixel on the image.2.A new weighting factor is designed to define the geodesic distance.The new weight is defined as the probability density estimation of gradient value.A new gradient calculation method is proposed,naming the seeding point sensitive gradient based on probability density.The probability density function is integrated to make the contrast between different tissues more obvious and the process of gradient calculating more reasonable.The newly defined geodetic distance is called the probability density-weight based geodetic distance,used as the similarity measure of the pixel point for the preliminary of superpixel segmentation.3.Increase the post-processing of local segmentation,to partial segment the superpixel,making the segmentation more accurate,improve the precision of the classification of superpixel.Even with the most classical FCM(Fuzzy C-Means Clustering)algorithm,there is still a very good effect.Finally combining probability density weighted geodetic distance based superpixel segmentation and FCM,we get the segmentation of brain tissue segmentation on the basis of superpixel.Experimental results show the superpixel generated by the algorithm has a more precious segmentation boundary than others,can accurately distinguish each brain tissue,and further improve the accuracy of image segmentation.The medical image segmentation,that is,the brain MR image segmentation technology is applied to the medical project:3D visualization system of cranial brain MR images.Cranial brain tissue reconstruction is performed on the basis of segmentation,and the 3D geometric model of the brain is constructed.The reconstructed 3D image can vividly display the structural view of the brain,can help doctors diagnose,understand the structure of the brain circuit,analyze the neural circuits and diseased tissues.