Research And Implementation Of Medical Image Segmentation Algorithm

With the development of medical imaging technology and computer science, Medical image processing technology becomes an important medical auxiliary technology of modern clinical diagnosis. Especially medical image segmentation is one of the most important steps before image analysis. However, the accuracy of segmentation and real time application could not come up to clinical requirements in the result of the complexity of tissue organ, the diversity of imaging mechanism and the high dependency of artificial interaction. Therefore, the accurate, fast and low artificial dependent segmentation studies become the research attention.In this thesis, the main research primarily focuses on normal tissue and lesions segmentation. The main works are as follows:1. A novel algorithm formulated by modifying the objective function in the fuzzy C-means algorithm to include a bias field which is modeled as a linear combination of a set of basis functions is proposed. Bias field estimation and image segmentation are simultaneously achieved as the result of minimizing this modified fuzzy C-means objective function. The efficient iterative algorithm for objective function minimization we provide converges to the optimal solution at a fast rate. The method has the superior performance compared with other methods.2. The traditional medical image segmentation algorithm can’t deal with intensity inhomogeneities. A region scale fitting level set model that draws upon intensity information in local regions at a controllable scale is proposed to overcome the difficulties. In addition, the regularity of the level set function is preserved by the level set regularization term to ensure accurate computation and avoids reinitialization expense. However, it is simply implemented by the full domain algorithm, in which the computation is carried out on the entire image domain. There is still plenty of room for improvement in computational and accurate efficiency.3. In order to deal with the segmentation of the interested region, such as brain tumors and other lesions, a modified region-scalable fitting (RSF) model and a more efficient narrow band algorithm has been presented to perform level set evolution. A new distance regularization term is used to maintain the regularity of the level set function, which is necessary for maintaining stable level set evolution and ensuring accurate numerical computation. The computational efficiency of our algorithm is further improved by using1D directional convolutions to approximate the2D convolutions in the computation of the two fitting functions in the RSF model. Our algorithm has been tested on synthetic and real medical images with promising results.4. The modified FCM algorithms are realized in3D space and directly applied on3D volume data, which supply the experimental data for realize3D reconstruction and visualization.