摘要
针对卷积神经网络在处理图像配准中没有考虑到特征重要性以及局部空间变换能力不足等问题,提出一种基于特征块(feature patch)的双通道局部空间变换网络用于医学图像配准。基于双路编码-解码网络,分别对固定图像以及浮动图像进行特征提取;基于特征金字塔,分别选取三层特征进行基于特征块的通道加权和双通道局部空间变换;用Dice系数衡量配准图像与固定图像的配准精度。在多个公开的脑部和肝脏数据集上的配准结果表明,该方法配准效果好,能有效提高配准精度。
The convolutional neural network does not take into account the importance of features and insufficient local spatial transformation capabilities in image registration.To solve this problem,we propose a dual-channel local spatial transform network based on feature patches for medical image registration.On the basis of the dual encoding-decoding network,feature extraction was performed on fixed images and moving images respectively.According to the feature pyramids,three layers of features were selected for channel weighting based on feature patches and dual-channel local spatial transformation.We used Dice coefficient to measure the registration accuracy of the registered image and the fixed image.The registration results on multiple published brain and liver datasets show that the proposed method has better registration effect and can effectively improve the registration accuracy.
作者
张纠
刘晓芳
杨兵
Zhang Jiu;Liu Xiaofang;Yang bing(Institute of Computer Application and Technology,China Jiliang University,Hangzhou 310018,Zhejiang,China;Institute of Electronic Information and Communication,China Jiliang University,Hangzhou 310018,Zhejiang,China;Key Laboratory of Electromagnetic Wave Information Technology and Metrology of Zhejiang Province,College of Information Engineering,China Jiliang University,Hangzhou 310018,Zhejiang,China)
出处
《计算机应用与软件》
北大核心
2022年第6期148-154,175,共8页
Computer Applications and Software
基金
国家自然科学基金项目(61672476)
浙江省大学生科研创新活动计划项目(2019R409055)。
关键词
图像配准
特征重要性
特征块
空间变换
特征加权
Image registration
Feature importance
Feature patch
Spatial transform
Feature weighting
