Medical Image Segmentation Based On Active Contour Method

The role of medical imaging in the diagnosis, treatment of disease and some other fields has become more and more important. Active contour models are widely implemented in medical imaging segmentation area in recent years. These powerful models have proven to be effective in segmentation, visualization, match and track by exploiting constrains derived from the image data together. Active Contour Model could naturally convert complex segmentation into a functional optimization problem, which means the contour deforms according as defined energy minimization principle.Traditional active contour models are firstly introduced in this thesis, including basic theory, implementation, advantage and disadvantage. Gradient Vector Flow (GVF), which is a kind of representative active contour model, is secondly explained. GVF derives from generalized force balance equation. A vector diffusion equation is employed to diffuse the gradient of an edge map in regions distant from the boundary. Active contour model has its own disadvantages: First, it often subjects to initial contour, noise and fake edges. Second, it also can't syllabify segment the concave edge in an image.According those disadvantages mentioned above, this thesis offers two improvements. First, the smoothing item of GVF force is improved. Usually, the coefficient of the smoothing item of GVF force is constant, and practically we don't know whether the coefficient we set is comfortable or not. The improved method binds the coefficient and the gradient of the edge map, which could make the edge feature stronger. Second, the data item of GVF force is improved. Here a monotonously degressive function is adopted (in this thesis we use Sigmoid function).We set the Sigmoid function as the coefficient. In the first few steps, this function could indicate the area where the gradient of edge map is large. As the iterative goes on, the effect of the function decreases. So the final force is determined by the first item.There are some other improvements as follows: First, the unsatisfactory contour, which is usually near the concavity area, is cut off and here we provide a new contour which could be yielded by GVF Snake or other methods, then we link the two contours into a complete contour. This man-machine interaction could let medicinal experts use their experience in these medical analysis works. Second, redundant data processing is adopted. These algorithms are implemented by Matlab. The precision of Matlab is high, but this high precision is unnecessary for coordination of the contour. Before a new iteration, redundant data processing is done to induce computing time.