计算机科学

计算机科学 ›› 2025, Vol. 52 ›› Issue (7): 151-160.doi: 10.11896/jsjkx.240400159

• 计算机图形学&多媒体 • 上一篇    下一篇

EFormer:基于分频和广注意力的高效Transformer医学图像配准模型

黄星宇, 王丽会, 唐堃, 程欣宇, 张健, 叶晨   

  1. 贵州大学计算机科学与技术学院 贵阳 550025
    公共大数据国家重点实验室 贵阳 550025
    贵州省智能医学图像分析与精准诊断重点实验室 贵阳 550025
    文本计算与认知智能教育部工程研究中心 贵阳 550025
  • 收稿日期:2024-04-22 修回日期:2024-10-17 发布日期:2025-07-17
  • 通讯作者: 王丽会(lhwang2@gzu.edu.cn)
  • 作者简介:(gs.xingyuhuang21@gzu.edu.cn)
  • 基金资助:
    国家自然科学基金(62161004);贵州省科学技术基金重点项目(黔科合基础-ZK[2021]重点002);科学计划基金(黔科合基础-ZK[2022]一般046)

EFormer:Efficient Transformer for Medical Image Registration Based on Frequency Division and Board Attention

HUANG Xingyu, WANG Lihui, TANG Kun, CHENG Xinyu, ZHANG Jian, YE Chen   

  1. School of Computer Science and Technology, Guizhou University, Guiyang 550025, China
    State Key Laboratory of Public Big Data, Guiyang 550025, China
    Guizhou Key Laboratory of Intelligent Medical Image Analysis and Precision Diagnosis, Guiyang 550025, China
    Ministry of Education Engineering Research Center of Text Computing and Cognitive Intelligence, Guiyang 550025, China
  • Received:2024-04-22 Revised:2024-10-17 Published:2025-07-17
  • About author:HUANG Xingyu,born in 1999,postgraduate.His main research interests include medical image processing and medical image registration.
    WANG Lihui,born in 1982,professor,Ph.D supervisor.Her main research interests include medical imaging,medical image processing,machine learning and pattern recognition.
  • Supported by:
    National Natural Science Foundation of China(62161004),Guizhou Provincial Science and Technology Projects(QianKeHe ZK [2021] Key 002) and Guizhou Provincial Science and Technology Projects(QianKeHe ZK [2022] 046).

PDF (PC)

19

摘要： 医学图像配准对于多种后处理步骤至关重要。目前基于卷积和Transformer的单流或双流网络架构能够实现良好的配准性能,但在配准性能与计算效率之间仍然难以取得平衡。为了解决这个问题,提出了一种高效的Transformer配准网络EFormer。其主要由分频器模块(Frequency Division Module,FDM)和广注意力模块(Broad Attention Module,BAM)组成。具体而言,在编解码器中使用多个FDM模拟双流网络并行提取局部-全局信息以提高计算效率;利用BAM增强多个FDM中局部信息的传递以保留配准中重要的语义特征。在3个数据集上的定性和定量比较实验结果表明,相比主流配准模型,EFormer在DSC,ASSD,HD95和雅可比行列式负值百分比4个评价指标上分别至少优化了1.3％,2.6％,0.6％和95％。此外,使用EFormer-tiny时,计算效率(Flops)优化了14％,表明EFormer能够以最快的计算速度在基于Transformer的网络中实现最佳的配准结果。

关键词: 医学图像配准, 分频, 广注意力, 高效Transformer

Abstract: Medical image registration is essential for several post-processings.Even though the existing single-stream or dual-stream network structures based on convolution and Transformer can achieve the promising results,it is still difficult to make a compromise between the registration performance and computational efficiency.To deal with this issue,this paper proposes an efficient registration network EFormer which mainly consists of the frequency division module(FDM) and broad attention module(BAM).Specifically,stacking several FDMs in encoder and decoder to mimic the roles of dual-branch network for extracting both local and global information can significantly improve the computation efficiency,using BAM to enhance the transmission of local information in multiple FDMs can preserve significant semantic features to promote the registration performance.The qualitative and quantitative comparisons with state-of-the-art methods on three datasets demonstrate that the Dice score,ASSD,HD95 and the ratio of negative Jacobian determinant of the proposed EFormer is improved at least by 1.3%,2.6%,0.6% and 95% respectively.In addition,using EFormer-tiny,the computation efficiency(Flops) is improved by 14%,showing that the proposed EFormer can achieve the best registration results in attention-based networks with the fastest computation speed.

Key words: Medical image registration, Frequency division, Broad attention, Efficient Transformer

中图分类号: 

  • TP389.1
[1]VEIGA C,JANSSENS G,TENG C L.First Clinical Investigation of Cone Beam Computed Tomography and Deformable Re-gistration for Adaptive Proton Therapy for Lung Cancer[J].International Journal of Radiation Oncology Biology Physics,2016,95(1):549-559.
[2]NAKAO M,KOBAYASHI K,TOKUNO J.Deformation analysis of surface and bronchial structures in intraoperative pneumothorax using deformable mesh registration[J].Medical Image Analysis,2021,73:102181.
[3]ALVAREZ P,ROUZÉ S,MIGA M I.A hybrid,image-based and biomechanics-based registration approach to markerless intra-operative nodule localization during video-assisted thoracoscopic surgery[J].Medical Image Analysis,2021,69:101983.
[4]VERCAUTEREN T,PENNEC X,PERCHANT A.Diffeomorphic demons:Efficient non-parametric image registration[J].NeuroImage,2009,45(1):S61-S72.
[5]SHEN D G,DAVATZIKOS C.HAMMER:hierarchical attri-bute matching mechanism for elastic registration[J].IEEE Transactions on Medical Imaging,2002,21(11):1421-1439.
[6]AVANTS B B,EPSTEIN C L,GROSSMAN M.Symmetric diffeomorphic image registration with cross-correlation:Evaluating automated labeling of elderly and neurodegenerative brain[J].Medical Image Analysis,2008,12(1):26-41.
[7]BEG M F,MILLER M I,TROUVÉ A.Computing Large De-formation Metric Mappings via Geodesic Flows of Diffeomorphisms[J].International Journal of Computer Vision,2005,61(2):139-157.
[8]HEINRICH H P,JENKINSON M,BRADY M.MRF-Based Deformable Registration and Ventilation Estimation of Lung CT[J].IEEE Transactions on Medical Imaging,2013,32(7):1239-1248.
[9]KLEIN S,STARING M,MURPHY K.elastix:A Toolbox forIntensity-Based Medical Image Registration[J].IEEE Transactions on Medical Imaging,2010,29(1):196-205.
[10]HELLIER P,ASHBURNER J,COROUGE I.Inter-subject Re-gistration of Functional and Anatomical Data Using SPM[M]//Berlin:Springer,2002:590-597.
[11]MAES F,COLLIGNON A,VANDERMEULEN D.Multi-modality Image Registration by Maximization of Mutual Information[J].IEEE Transactions on Medical Imaging,1997,16(2):187-198.
[12]DE VOS B D,BERENDSEN F F,VIERGEVER M A.End-to-End Unsupervised Deformable Image Registration with a Con-volutional Neural Network[C]//LNIP.2017:204-212.
[13]JADERBERG M,SIMONYAN K,ZISSERMAN A.SpatialTransformer Networks[C]//Proceedings of the 29 International Conference on Neural Information Processing System.2015:2017-2025.
[14]BALAKRISHNAN G,ZHAO A,SABUNCU M R.An Unsupervised Learning Model for Deformable Medical Image Registration[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.2018:9252-9260.
[15]CHEN X,RAVIKUMAR N,XIA Y.A Deep Discontinuity-Preserving Image Registration Network[J].arXiv:2107.04440,2021.
[16]ZHENG Z.Feature self-calibration network with global-localtraining strategy for multi-region deformable medical image re-gistration[J].Neural Computing and Applications,2022,34:17175-17191.
[17]CHE T,WANG X,ZHAO K.AMNet:Adaptive multi-level network for deformable registration of 3D brain MR images[J].Medical Image Analysis,2023,85:102740.
[18]ZHAO S,DONG Y,CHANG E.Recursive Cascaded Networks for Unsupervised Medical Image Registration[C]//2019 IEEE/CVF International Conference on Computer Vision.2019:10599-10609.
[19]KIM B.CycleMorph:Cycle consistent unsupervised deformable image registration[J].Medical Image Analysis,2021,71:102036.
[20]SHEN Y,YAN Y,SONG J,et al.Brain magnetic resonanceimage registration based on parallel lightweight convolution and multi-scale fusion[J].Journal of Biomedical Engineering,2024,41(2):213-219.
[21]JIA X,BARTLETT J,ZHANG T.U-Net vs Transformer:Is U-Net Outdated in Medical Image Registration?[J].arXiv:2208.04939,2022.
[22]ZHOU Z,SIDDIQUEE M M R,TAJBAKHSH N.UNet++:Redesigning Skip Connections to Exploit Multiscale Features in Image Segmentation[J].IEEE Transactions on Medical Imaging,2020,39(6):1856-1867.
[23]CHEN J,HE Y,FREY E C.ViT-V-Net:Vision Transformer for Unsupervised Volumetric Medical Image Registration[J].ar-Xiv:2104.06468,2021.
[24]CHEN J,FREY E C,HE Y.TransMorph:Transformer for unsupervised medical image registration[J].arXiv:2111.10480,2022.
[25]LIU Z,LIN Y,CAO Y.Swin Transformer:Hierarchical Vision Transformer using Shifted Windows[J].arXiv:2103.14030,2021.
[26]MOK T C W,CHUNG A C S.Fast Symmetric Diffeomorphic Image Registration with Convolutional Neural Networks[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition.2020:4644-4653.
[27]MA M,SONG L,XU Y.Symmetric Transformer-based Net-work for Unsupervised Image Registration[J].arXiv:2204.13575,2022.
[28]WANG N N,CHENG Y Z.Cascaded Block Registration Model Based on Multi-spatial Information Extraction[J].Computer Systems & Applications,2023,33(2):125-133.
[29]ZHAO X,LI X J,XU J,et al.Parallel medical image registration model based on convolutional neural network and Transformer[J].Journal of Computer Applications,2024(12):3915-3921.
[30]HU X,KANG M,HUANG W.Dual-Stream Pyramid Registration Network[J].arXiv:1909.11966,2019.
[31]SUN Y,MOELKER A,NIESSEN W J.Towards Robust CT-Ultrasound Registration Using Deep Learning Methods[C]//Understanding and Interpreting Machine Learning in Medical Image Computing Applications.Cham:Springer,2018:43-51.
[32]CHEN X,ZHOU B,XIE H.Dual-Branch Squeeze-Fusion-Excitation Module for Cross-Modality Registration of Cardiac SPECT and CT[J].arXiv:2206.05278,2022.
[33]MA M,LIU G,SONG L.SEN-FCB:an unsupervised twinning neural network for image registration[J].Applied Intelligence,2023,53(10):12198-12209.
[34]FAN X,ZHUANG S,ZHUANG Z.SearchMorph:Multi-scale Correlation Iterative Network for Deformable Registration[J].arXiv:2206.13076,2022.
[35]FISCHL B.FreeSurfer[J].NeuroImage,2012,62(2):774-781.
[36]HERING A,HANSEN L,MOK T C W.Learn2Reg:compre-hensive multi-task medical image registration challenge,dataset and evaluation in the era of deep learning[J].IEEE Transactions on Medical Imaging,2022,42(3):697-712.
[37]PASZKE A,GROSS S,MASSA F.PyTorch:An ImperativeStyle,High-Performance Deep Learning Library[C]//NeurIPS 2019.2019.
[38]LUO W,LI Y,URTASUN R.Understanding the Effective Receptive Field in Deep Convolutional Neural Networks[C]//Proceedings of the 30th International Conference on Neural Information Processing Systems.2016:4905-4913.
[39]NAKKIRAN P,KAPLUN G,BANSAL Y.Deep double de-scent:where bigger models and more data hurt*[J].Journal of Statistical Mechanics:Theory and Experiment,2021,2021(12):124003.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
渝ICP备16013121号-4
地址：重庆市渝北区洪湖西路18号    邮编：401121  
电话：023-63500828（编辑部） 023-67039612（会议/专题） 023-67039623（财务/发票）
E-mail：jsjkx12@163.com
本系统由北京玛格泰克科技发展有限公司设计开发