基于图割与概率图谱的肝脏自动分割研究

Automatic Liver Segmentation Using Graph Cuts and Probabilistic Atlas

提出一种利用均值漂移算法做预处理,将概率图谱与图割算法相结合的肝脏自动分割方法。该方法一方面有效利用概率图谱所代表的形状信息,并且以无参形状建模;另一方面,均值漂移算法对灰度和形状信息组成的综合信息进行过分割处理,利用过分割的区域代替单个像素参与图割算法,降低了优化算法的复杂度。实验结果表明,该方法有效结合了概率图谱和均值漂移算法的优点,提高了图割算法的精度和速度。

The paper proposed a method for automatic liver segmentation.Firstly mean shift algorithm was used in a preprocess step,and then graph cuts and probabilistic atlas algorithm were combined to segment the liver.This method has some advantages,for one thing,it makes good use of the shape information contained in probabilistic atlas and pro-babilistic atlas method is a non-parametric model,for another,mean shift algorithm processes the composite information,including gray value and shape index,and then pixels are replaced by super-pixel to reduce the computational complexities of graph cuts algorithm.Experiment results show that the proposed method possesses the nice properties of the mean shift and probabilistic atlas method,and is both efficient and accurate.