摘要
针对SOM网络分割图像,随着神经元数量增加,分割性能变差且无法分割噪声强度过大图像的问题,提出一种用于医学图像分割的基于神经网络的方法。提出一个改进的自组织映射(SOM)网络,FIR-SOM网络将有限脉冲响应(FIR)加入SOM中,把每个神经元作为FIR系统,在FIR-SOM方法分割后,通过合并聚类法把联合聚类的对象连接在一起。乳腺超声检查图像(BUS)的分割结果表明,该算法能够有效分割肿瘤区域,分割效果优于基于SOM的方法。
Self-organizing map（SOM）network does not usually result in better segmentation performance with the increasing number of neurons.And it can not segment images with heavy noise successfully.A neural network based method for medical image segmentation was proposed.A modified self-organizing map（SOM）network was proposed.A finite impulse response（FIR）was added to the SOM network.In this network,each neuron acted as a finite impulse response（FIR）system.After the preliminary segmentation,a merging process was ready to connect the objects of a joint cluster together.The segmentation results of breast ultrasound images（BUS）show that the proposed method can segment tumor region effectively and it works better than SOM-based methods.
出处
《计算机工程与设计》
北大核心
2016年第9期2533-2537,F0003,共6页
Computer Engineering and Design
基金
十一五国家科技支撑计划基金项目(2010BAI88B00)
关键词
人工神经网络
医学图像分割
模式识别
合并聚类
自组织映射
artificial neural network（ANN）
medical image segmentation
pattern recognition
combined clustering
self-organizing map
