Femur DR Image Segmentation Based On The Variational Level Set Method

The medical images play an increasingly important role in modern medicine.Doctors can quickly diagnose the illness and make a surgical plan with the help ofan accurate understanding of medical images. In recent years, medical imagesegmentation algorithm has become a research focus. The main task is to solve theproblem of intens ity inhomogene ity. As the object of our research, femur DRimages belong to one kind of medical images. DR images of the femur in vivohave the characteristics of the uncertain target numbers and intens ityinhomogene ity, e.g. Traditional segmentation algor ithm is able to extract the femurcontours, however, the characteristics of the fracture cross-section extraction isvery difficult.In accordance with the concept of the gray-scale fluctuations, in this paper，we analyzed the gray-scale fluctuation curve in the horizontal direction in the DRimages. In this way, we gave a more explic it explanation of the intens ityinhomogene ity of the image. Based on the C-V model, combining with gray-scalefluctuations, we got a novel leve l set image segmentation model, which is effectivefor fe mur DR image segmentation. In this dissertation, we also analyzed andsummarized a large number of image segmentation algorithm proposed by scholarsat home and abroad.Firstly, we detailed the basic princ iples and concepts of curve evolution theoryand the variational leve l set method, discussing some very extensive and widelyinfluentia l classic models: MS model, the C-V model, RSF model. We gave thepros and cons of them by a discussion of model physics, and the significance ofthe energy functional. Secondly, the gray-scale fluctuation theory was used toanalysis the intens ity inho mogeneity of the image. We used B-spline to fit thegray-scale fluctuation curve and got a smoother one. And then, combining CVmodel which is one of the level set algorithms and adding a penalty term makesthe functional won’t need to re-initialize, we propose a novel segmentation modelthat has a capacity to handle images with intensity inhomogene ity. The model usesinformation not only statistics of the whole image but also gray-scale fluctuations.Besides the robust to the initial curve, it has good noise immunity. By adding apenalty term, without re-initia lize the level set function, the model segmentationspeed was greatly accelerated. In order to verify the algorithm validity, we tested itin the blood vessel X-Ray image and got a very good segmentation results. Then,the dissertation described the segmentation process of the femur DR image in vivo.Finally, the medical image segmentation quality evaluation methods have beenintroduced. Respective ly, we used the subjective method and objective method toevaluate the effects of image segmentation.