Medical Image Segmentation Based On Active Contour Model

The medical image mainly includes computerized tomography (CT), ultrasound image and magnetic resonance (MR) image, etc. It plays an important role in fields of quantitive analysis, real-time monitoring and treatment scheduling, etc. As medical images are utilized to do medical analysis and diagnosis, the specific organs or regions of interest (ROI) are usually extracted, so as to du further analysis better. The process is image segmentation. The general segmentation problem is the process of partitioning an image or data-set into a number of uniformity or homogeneous segments, and the incorporation of any two adjacent segments will break the uniformity or homogeneity. Medical image segmentation is important for medical image analysis, for example,3D Visualization, Computer Aided Operation, and Radiology Treatment all assume that Region of Interesting (ROI) are well segmented. Traditional image segmentation methods can not satisfy us because mescal images have low resolution, low contrast and intrinsic noises.Active contour model, which has been used widespread, is a new image segmentation and tracking method that is researched in recent ten years. It researches the evolvement process of the curve in dynamics angle, and gets continuous and close borders of the image regions of interest by computation of the minimum of the energy function. Because active contour model is easy for modeling, as well as can extract objects contours which have any shape, it has widespread applications and great development in fields of medical image segmentation, edge detection and motion tracking, etc. It also has efficient and simple calculation. Practices prove that this method has great improvement than former image segmentation methods and adapts to medical image segmentation. But it is so new to be restricted to use in clinical practice in some degree because of problems itself. Based on the above facts, this paper does all-around researches on active contour model, in order to improve current algorithms or propose new methods for medical image segmentation. Therefore, this thesis performs in-depth and systematic studies on the topic named by medical image segmentation based on active contour model.This thesis first gives an overview of medical image segmentation methods, and analyzes the advantages and disadvantages of these methods. Two large kinds of active contour models, that is, parametric active contour models and geometric active contour models, and level set method are then discussed and analyzed. Chan-Vese (CV) model and its applications in heterogeneous images especially medical image are focused. Some drawbacks in certain aspect are found from the analysis, and then the model has been improved in this thesis. The works of this paper are show as follow:1. CV model based on Kullback-Leibler (KL) distance (KLCV model) is proposed for the problem as parameters of CV model are difficult to choose in segmentations of heterogeneous images. Because KL distances change along with the change of curve, parameters can automatically adjust to homogeneity balance in the energy function. In addition, the statistics information is embedded to the traditional CV model which hence segments medical images no longer based on only global regions information of images. All works above are in order to enhance the control of the curve evolution.2. There are some problems such as heterogeneous gray level, complex construction of tissues and organs, etc. in medical images. Therefore in the field of medical image segmentation, traditional CV model and KLCV model can get more desirable results, as the reason that they are only based on global image information. The KL distance weighted CV model based on local neighborhood information is proposed on the basis of above models. In this new model, the analysis of an image leads to the construction of a family of local neighborhood energies at each point along the curve. This work causes that evolving curve will detect the variation gray level in a small range, which results in that the heterogeneous regions far away the evolving curve will not influence the curve evolution. All works above are in order to optimize the segmentation performances of the new model in heterogeneous image segmentations.3. This thesis validates the new model with experiments on plenty of clinical medical images. The segmentation results show that the new model can segment accurately the objects from images, and the evolving curves have fine stability.