摘要
医学图像配准在医学图像分析领域扮演着至关重要的角色.众多学者已经在医学图像配准方面展开深度学习技术的算法研究,以取得比传统方法更好的性能.然而,绝大多数算法更专注于图像的空域信息,忽视了频域信息的潜在作用.本文提出了一种基于傅里叶神经算子(Fourier Neural Operator,FNO)的算法,利用傅里叶变换将图像转化到频域空间,在频域空间内学习图像特征,结合空域和频域的信息,利用深度网络模型能更全面地提取图像的各种特征.该算法经在ACDC和OASIS两个数据集上进行验证,均能够在保证配准速度的前提下,得到很好的配准效果.同时,本文对FNO模块进行了详细评估.
Medical image registration plays a crucial role in the field of medical image analysis.Deep learning technologies have been introduced to enhance the performance of medical image registration compared to traditional methods.However,most of the algorithms only focus on the spatial-domain information of the image,ignoring the potential role of the frequency-domain information.This paper proposes an algorithm based on the Fourier Neural Operator(FNO),which utilizes Fourier transform to convert images into the frequency domain,learns image features in the frequency domain,combines the information in both the spatial and frequency domains,and extracts the various features of the image in a more comprehensive way using the deep network model.The algorithm has been verified on ACDC and OASIS datasets,and both of them are able to obtain good registration results under the premise of guaranteeing the registration speed.Additionally,this paper provides a detailed evaluation of the FNO module.
作者
黄沛明
陆飞
方路平
HUANG Peiming;LU Fei;FANG Luping(College of Information Engineering,Zhejiang University of Technology,Hangzhou 310023,China)
出处
《小型微型计算机系统》
北大核心
2025年第6期1409-1415,共7页
Journal of Chinese Computer Systems
基金
浙江省重点研发计划项目(2020C03073)资助.
关键词
医学图像配准
神经算子
深度学习
无监督学习
medical image registration
neural operator
deep learning
unsupervised learning
作者简介
黄沛明,男,1997年生,硕士研究生,研究方向为医学图像配准,E-mail:2112003016@zjut.edu.cn;陆飞,女,1979年生,博士,讲师,研究方向为信号处理、人工神经网络、多维空间理论分析;方路平,男,1968年生,硕士,教授,CCF会员,研究方向为深度学习、医学计算.
