一种基于生物仿真结构的神经网络在超声医学图像分割中的应用  

Application of A Biologically Inspired Oscillator Network in Ultrasound Medical Image Segmentation

作　　者：陈冠饶[1] 王景熙[1] 郑昌琼[2] 王章伟[2] ZHENG Yi 

机构地区：[1]四川大学计算机学院,四川成都610065 [2]四川大学生物医学工程中心,四川成都610065 [3]Dept of Electrical Eng. Minnesota State Univ.,St.Cloud

出　　处：《四川大学学报（工程科学版）》2001年第3期113-116,共4页Journal of Sichuan University (Engineering Science Edition)

基　　金：国家自然科学基金资助项目! (6 96 310 2 0 )

摘　　要：提出将LEGION(LocallyExcitatoryGloballyInhibitatoryOscillatorNetwork)这种新的基于生物仿真结构的神经网络应用于分割心脏的超声图像 ,讨论了LEGION中参数对分割结果的影响。由于LEGION具有天然的并行、分布式处理的特性 ,与一些现有的分割方法进行了比较 ,结果表明本文所用的方法所得的分割图像边缘更为清晰具体 ,细节保留更为良好。Image segmentation has always been one of the difficult problems in image processing. It is the key from image processing to image analysis and the first step of automatic image analysis. Ultrasound images of human heart, because of its noise and fuzzy boundary, are especially hard to segment. In this paper, a biologically simulated structure, LEGION (Locally Excitatory Globally Inhibition Oscillator Network), is first introduced and then applied to segment the sample images. Parameters of the algorithm are analyzed. And at last, the result is compared with those produced by some other methods.Due to its feature of parallel and distributed processing,images processed by LEGION generally have clearer boundary and better details than those segmented by some traditional methods.

关 键 词：超声心脏图像 图像分割 LEGION 生物仿真结构 神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] R445.1[自动化与计算机技术—计算机科学与技术]