Study Of Interactive Segmentation Of Medical Image Sequences

Image segmentation is one of the most critical processing steps between image acquisition and image analysis. It aims to extract the region of interest from complex scenes. For the medical image segmentation, a good accuracy of results is very important and helpful for doctors to diagnose the illness and make the right therapeutic schemes. However, the current segmentation results which are extracted by the computer itself or lots of manual intervention are not accepted for the bad accuracy in medical practical applications. Therefore, the interactive segmentation studies with less manual intervention have become the research focus of medical image segmentation.In this thesis, the interactive segmentation based on medical image series is primarily studied and corresponding solutions for the problems are proposed. Experiments have verified the effectiveness of the solutions. The main works are as follows:1. For the lack of the traditional Live Wire interactive segmentation algorithm, it is improved in both the cost function and the optimal path search. With the correct location of interaction points, this improved Live Wire algorithm will be able to quickly extract the accurate edge information from the interested object.2. A new interactive segmentation method based on Canny edge detection algorithm is proposed. Firstly, describes the implementation principle of Canny operator which is improved by adaptive dual-threshold calculation. Using of interactive operations, the closed edge of interested object will be extracted by the methods of searching breakpoints, connecting breakpoints, removing burr and so on. Finally, coupled with the morphology expansion method, the closed edge in the adjacent slice can be obtained in the same way automatically. It is verified by the experiments results.3. Another new interactive segmentation method based on GVF improved Snake model with the initial contour is proposed. Firstly, input a line segment intersecting the interested object edge by user interaction. Then extract the initial contour automatically. The GVF is improved by histogram enhancement. And adjust the contour points location to make sure that they are evenly distributed between the deformation process. The active contour will converge to the edges of interested object successfully. Finally, the medical image series will be processed in the same way by set the segmentation result of last slice as initial contour.4. Experiment show that all of these three kinds of improved segmentation algorithm are able to achieve the accurate segmentation for a single image. However, the segmentation results of Live Wire algorithm will deviate from the true edge position, if the the location of interaction points is bad. And although the results from Canny operator are the most accurate, it may not be able to extract the complete closed edge of the target because of the noise. For the medical image series segmentation, the GVF improved Snake model algorithm is the best to extract the closed boundary of interested object quickly and reliably with only little user intervention.