机构地区: 滨州医学院卫生管理学院
出 处: 《安徽大学学报(自然科学版)》 2014年第2期33-39,共7页
摘 要: 为了更好地分割医学图像,对传统的神经网络进行改进,对分割后的图像区域特征进行约减,以降低特征向量维数,同时抽取出规则,根据这些规则构造神经网络隐含层的神经元个数,确定神经网络的初始拓扑结构.然后用逆推学习算法迭代,得到最终的决策结果,即实现图像的分割.实验证明,该方法大大缩短了实验时间,提高了精度,并且得到优于常规的分割图像,满足图像处理的事实性要求. In order to segment medical images more efficiently,the research improved the traditional neural network,subtracted the image regional characteristics after segmentation,so as to lower feature vector dimension.This study was based on the extracted rules the number of neurons in hidden layer neural network neurons,and determined the initial neural network topology.A reverse iterative learning algorithm was utilized in this study; the final decision results were achieved and the image segmentation was realized.The experiments showed that this method greatly reduced the experiment time and improved the accuracy.The segmentation image was superior to conventional image segmentation and satisfied the factual requirements of the medical image processing.