基于水平集方法的胎儿3维超声图像交互分割算法  

Interactive Segmentation Method of 3D Fetal Ultrasound Image Based on Level Set Approach

作　　者：李昕[1,2] 付忠良[1,2] 张丹普[1,2] 

机构地区：[1]中国科学院成都计算机应用研究所,四川成都610041 [2]中国科学院大学,北京100049

出　　处：《四川大学学报（工程科学版）》2013年第3期85-90,共6页Journal of Sichuan University (Engineering Science Edition)

基　　金：四川省科技支撑计划基金项目(2011GZ0171;2012GZ0106)

摘　　要：针对目前胎儿3维超声成像主要靠经验丰富的医生手动方式来分割母体部分的不足,提出了一种基于水平集演化的3维超声图像交互分割方法。该算法通过构造基于梯度方向和大小的演化控制函数来找到处于零水平集的物体边界,为了将水平集方法应用于3维超声,提出窄带方案将演化过程限定在与初始态曲线等距的2条曲线中,而不是在每次演化中更新3维图像中的每个点,提高了分割的速度和精度。实验中使用2种不同来源的超声图像集,它们是来自于不同位置、不同方向、不同时期的胎儿30幅图像。结果显示,该方法能对3维胎儿超声图像进行快速准确的分割,在处理器为dualcore 2.0 GHz的机器上能在5 s内完成对像素大小为255×255×128的图像进行分割。In order to overcome drawbacks of manually segmenting the maternal artifacts from 3D ietal uhrasound image even by experienced sonographers, an interactive segmenlation approach was proposed based on level set flamework. This algorithm could deleet the boundary of objects by controlling the propagation speed function based on gradient orientation anti scalar. In o,＇der to apply level set into 3D ultrasound segmentation, the Narrowband scheme was put forwarcl, which increases the segmenting speed and accu racy dramatically by restricting propagation pcess between two equidistant emwes instead of updating it at all the on the grid. Experiments were performed on 3D felal uhrasotmd datasets from two different uhrasound machines containing 30 fetal images of differ ent position, orientati,m and weeks. The resuh showed that this approach achieves fast and accurate segmentation of 3D fetal uhrasound image with size ,,f 255 x 255 x 128 in 5 seconds on a dualcore 2.0 GHz computer.

关 键 词：3维胎儿超声 水平集 图像分割 窄带方案 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]