摘要
利用心脏核磁共振成像技术对左心室进行分割,可以准确计算出心室容积等重要临床指标。针对左心室位置检测、形状推断与分割问题,提出一种基于卷积网络和可变模型算法的左心室图像处理方法。利用稀疏自动编码和卷积网络实现左心室图像位置的高精度检测;基于堆栈稀疏编码器和多层神经网络推断出左心室图像的基本形状;利用可变模型和推断出的形状组合对心脏图像进行精确分割。在30个心脏核磁共振数据库中采集图像数据进行实验分析,实验结果表明,相比其他几种较新的分割算法,该方法在计算轮廓比例和一致性两个指标上均获得了最优结果。
The segmentation of left ventricle by cardiac MRI can accurately calculate the important clinical indexes such as ventricular volume.Aiming at the problems of left ventricular position detection,shape inference and segmentation,this paper proposes a left ventricular image processing method based on convolution network and variable model algorithm.Sparse automatic coder and convolution network were adopted to detect the position precisely.Stack sparse encoder and multilayer neural network were used to infer the basic shape of left ventricular image.Using the combination of variable model and inferred shape,we segmented the heart image accurately.We carried out experiments on image data collected from 30 cardiac MRI databases.The experimental results show that compared with other new segmentation algorithms,our method obtains the optimal results in the calculation of contour ratio and consistency.
作者
任侠
胡玉平
Ren Xia;Hu Yuping(School of Information Engineering,Quzhou College of Technology,Quzhou 324000,Zhejiang,China;School of Information Science,Guangdong University of Finance and Economics,Guangzhou 510320,Guangdong,China)
出处
《计算机应用与软件》
北大核心
2020年第1期230-236,283,共8页
Computer Applications and Software
基金
广东省自然科学基金项目(2016A030313717)
浙江省2018年度高校国内访问学者专业发展项目(FG2018159)
2018年度衢州市科技计划指导性项目(2018014)
关键词
磁共振
左心室分割
卷积神经网络
稀疏编码器
可变模型
Magnetic resonance
Left ventricle segmentation
Convolutional Neural network
Sparse encoder
Variable model
