一种新颖的前列腺超声图像分割方法

Novel segmentation method for prostate TRUS image

现有的基于先验知识的超声图像分割方法往往都需要对先验形状进行建模,由于建模一般都是对形状的整体进行学习,造成最终分割结果往往在细节上会有所不足,并且在分割过程中都涉及到对形状的迭代,对分割时间也会造成一定的影响。为了克服现有先验知识的不足,提出一种新颖的图像分割方法。该方法提出了一种边界描述符,定义了分割过程中前后检测点的空间关系,使用该空间关系能够将边界点检测过程中出现的异常(噪声)点自动过滤掉,从而摒弃掉了对先验知识的依赖。同时,该描述符改进了传统的法向量对比度边界算子,考虑了其邻域法向量上的边界信息,使得边界检测性能得到了极大的提升。实验结果表明,使用该边界描述算子能够直接将前列腺的边界分割出来,相比传统的边界检测过程,迭代次数可以忽略不计,在时间和精度上都得到了极大的提升。

In terms of the existing ultrasound image segmentation methods based on prior knowledge,the prior shape needs to be modeled,usually.The modeling generally means learning the whole shape,so the final segmentation results are often inadequate in details and the shape iteration is involved in the segmentation process.In addition,the segmentation time will be impacted to a certain extent.In view of the above,a novel image segmentation method is proposed to overcome the deficiency of the prior knowledge.In this method,a boundary descriptor is proposed,which defines the spatial relationship between the front and rear detection points in segmentation process.With the spatial relationship,the abnormal(noise)points appearing in the detection process of boundary points can be automatically filtered out,so as to discard the dependence on prior knowledge.The descriptor improves the traditional normal vector contrast boundary operator and also greatly improves the performance of boundary detection because it takes the boundary information of its neighborhood normal vector into account.The experimental results show that the prostate boundary can be directly segmented by this boundary description operator.In comparison with the traditional boundary detection process,the iterations of this method can be ommited,its detection time is shortened and its detection accuracy is improved greatly.