基于能量驱动的分水岭算法在MRI海马图像分割中的应用

The Application of Energy-Driven Watershed Transform Algorithm to Hippocampus Segmentation in MRI

传统的分水岭算法虽然灵活性强,但在分割过程中较少用到边界特征的信息,使得图像的过分割问题比较突出.提出利用能量驱动的分水岭算法来实现人脑MRI中的海马分割.利用分水岭算法模型计算水流从种子点出发,沿代价最小的路径流经每一个像素的代价,将该像素的代价作为像素的能量.在整个图像能量最小化驱动下修改初始分水线处像素的归属类别,使分割结果与目标物体轮廓重合.该算法将内部特征与边缘条件相结合,可以很好地限制分水岭算法过分割的问题.多套MRI海马的分割结果表明,该算法可应用于海马等复杂结构的分割.

Traditional watershed transform algorithm is flexible, but some regional characteristics in the boundary are seldom used. So, over-segmentation is serious in segmentation results. A method is presented, which employs energy-driven watershed transform algorithm to segment hippocampus in human brain MRI. For this method, the watershed transform model is used. Water floods to each pixel along the shortest path from seed points, then computes the cost of flooding across the pixel. The cost can be used as the energy of this pixel. Driven by energy minimizing, the class to which the points in the contour belong is modified to make the contour develop into the ultimate result. Some inner characteristics and boundary condition are used to limit over-segmentation. The segmentation results show this method can be applied to the segmentation of some complex structures, such as hippocampus.