Fast Interactive Segmentation Of 3D Medical Images

Image segmentation is to merge the regions which has high similarity. Medical image segmentation has wide application such as quantitative analysis and detection, visualization and reconstruction of three-dimensional human body. It is important to segment the region of interest. Interactive segmentation method is common method for medical image segmentation which meet special segmentation task. With the need of rapid processing of large amounts of patient image data, doctors and researchers are desiring for real-time interactive segmentation.According to the actual needs of medical imaging applications, this paper presents a fast interactive segmentation method for three dimensional medical images. This paper uses a modified watershed method to cluster the image into super pixel. We purpose a multi-tree structure based on the super pixel according to the merge precedence that prefer merge the adjacent nodes with low relative gradient between the adjacent nodes and then come up with an efficient algorithm to divide multi tree into several subtree. Then we construct a graph structure and build the minimum spanning tree to realize the segmentation. The paper aims to eliminate the redundant edges of over-segmentation through the merging process of each layer.We test the method on variety category of three dimensional medical image of interactive segmentation such as abdominal computed tomography, brain computed tomography, brain magnetic resonance and so on. Experiments show that this method can achieve real-time interactive segmentation with minimal user interaction. The paper compares graph cut method, graph cut via tesson clustering method and the fast interactive segmentation method with respect to two aspects including running time and used interaction. The method can complete object segmentation at the milliseconds level and allow repetitive user interaction to refine the segmentation. The average rate of the segmentation accuracy is 80% and the level of boundary precision is in pixels.