基于多尺度位置感知网络的脑部MR图像配准  

Brain MR image registration based on multi-scale location-aware network

作　　者：杜晓刚 孙浩轩 王营博 雷涛[1,2] DU Xiao-gang;SUN Hao-xuan;WANG Ying-bo;LEI Tao(Shaanxi Joint Laboratory of Artificial Intelligence,Shaanxi University of Science and Technology,Xi’an 710021,China;School of Electronic Information and Artificial Intelligence,Shaanxi University of Science and Technology,Xi’an 710021,China)

机构地区：[1]陕西科技大学陕西省人工智能联合实验室,陕西西安710021 [2]陕西科技大学电子信息与人工智能学院,陕西西安710021

出　　处：《计算机工程与设计》2025年第3期879-885,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(61861024、62271296、62201334);陕西省教育厅科研计划基金项目(23JP022、23JP014);陕西省重点研发计划基金项目(2024GX-YBXM-121)。

摘　　要：为提高基于深度学习的可变形配准在图像复杂形变区域的配准精度,提出一种用于脑部MR图像配准的多尺度位置感知网络。设计一个位置感知注意力模块,分别沿3个坐标方向使用一维池化将位置信息编码生成特征图,定位图像的复杂形变区域;在编解码连接处使用多尺度策略,采用空洞卷积提取多尺度特征并融合,增强复杂形变特征的表达能力。在脑部MR数据集上进行配准实验,其结果表明,与主流方法相比,所提配准网络取得了更高的配准精度。To improve the registration accuracy of deformable registration based on deep learning in complex deformation areas of the image,a multi-scale location-aware network for brain MR image registration was proposed.A location-aware attention module was designed,in which the one-dimensional pooling along three coordinate directions was used to encode position information and generate feature maps,thus locating complex deformation areas in the image.A multi-scale strategy was employed at the encoder-decoder connection,in which the dilated convolution was used to extract and fuse multi-scale features,thus enhancing the ability to express complex deformation features.Experimental results on the brain MR datasets show that the proposed network achieves higher registration accuracy compared with the mainstream methods.

关 键 词：深度学习 可变形配准 无监督学习 位置感知 区域定位 特征融合 多尺度特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]