【摘要】 为了改善传统分割算法对初始条件的依赖性,提出一种基于自适应区域的医学图像自动分割框架.该框架将分割算法和目标检测技术集成到一起,通过检测已经分割出的目标信息来预测该目标在待分割切片上的局部区域,然后基于该局部区域预估目标的阈值范围和几何形状等信息,为应用分割算法提供初始参数.该框架已经应用到区域生长、基于阈值的水平集等常用分割算法.实验结果表明,该框架有效减少了分割中的人工交互操作,并且能自动处理目标复杂分叉情况.更多还原

【Abstract】 In order to decrease the dependency of the conventional segmentation algorithms on initial conditions,a medical image segmentation framework based on an adaptive region is presented. The framework integrates the segmentation and object detection techniques into a pipeline,and the local region of objects in the slice to be segmented is determined according to the detection results of the segmented slice. Based on this region,the threshold and geometric shape parameters can be obtained and used to the successive segmentation. This framework has been applied to region growing and threshold-based level set algorithms to segment lots of slice images. The experimental results show that the proposed framework can effectively decrease manual interaction,and well handle the complicated bifurcations of objects.更多还原