Research Of Medical Image Segmentation Algorithm Based On Geometric Active Contour Models

With the development of medical imaging technology, medical images have increasingly become an important basis for the clinical disease diagnosis. Traditional medical image diagnosis mainly relies on visual observation of the experience of doctors, subjectively and easily lead to misdiagnosis. Recently, medical diagnosis largely depends on image processing, especially that the research and development of surgery program planning and navigation system put forward higher requirements for medical image processing, such as the body tissues and organs segmentation, quantitative analysis, registration, three-dimensional reconstruction and visualization, and image segmentation is the foundation and key parts of the image processing. In this paper, the research is around image segmentation.The geometric active contour models based on curve evolution theory and level set method with the advantages of time effectiveness, the number of iterations, segmentation accuracy, robustness, stability and anti-noise, has been widely studied in the field of medical image segmentation. In this paper, combing the specific application, there are some research and improvement for the model, following are the major content:(1) The medical imaging principle, medical difficulty of images segmentation and theoretical foundation in mathematics of geometric active contour model have been analyzed and studied in detail; then deduce the relationship between curve level set evolution and image segmentation field. Finally, focus on functional design process of parametric active contour models and geometric active contour model.(2) Research mainly focuses on piecewise constant model based on region information and without re-initialize model based on edge information; As for the latter model sensitive to the initial contour, design a variable weight coefficients an internal energy term to propose an adaptive without re-initialize the model. Finally, we combine the advantages of noise immunity and robustness of regional model and time efficiency and accuracy of weak edge adaptive without re-initialization model, and the characteristics of medical image, propose an adaptive comprehensive model. The experimental results show that noisy medical image segmentation can get ideal results, while time efficiency is greatly improved.(3) This paper mainly analyzes the energy functional design idea of the segmentation effectiveness to the intensity inhomogeneity medical image of LBF and LRICV local fitting model, but the both of the quality of image segmentation are heavily dependent on the selection of the core radius, it needs to keep trying until the best results, otherwise it will easily lead to redundant too many false contour or miss potential weak edge, that is to say, it is a very subjective. In order to solve the problem, an edge keeping local fitting model is proposed by introducing a geodesic time kernel function which is combined the spatial distance information and image data information to redesign the energy function. This model is able to retain the potential weak boundary though the kernel radius big enough, adaptively select the neighborhood sampling point, to avoid the selection of kernel radius. The experimental results show that the proposed model has improved the performance of evolution curve into a local minimum and robustness of initial contour with greater accuracy and time efficiency of medical image segmentation.