基于变分推断的磁共振图像群组配准

doi:10.11938/cjmr20212918

摘要/Abstract

摘要：

为解决基于深度学习的成对配准方法精度低和传统配准算法耗时长的问题，本文提出一种基于变分推断的无监督端到端的群组配准以及基于局部归一化互相关（NCC）和先验的配准框架，该框架能够将多个图像配准到公共空间并有效地控制变形场的正则化，且不需要真实的变形场和参考图像．该方法得到的预估变形场可建模为概率生成模型，使用变分推断的方法求解；然后借助空间转换网络和损失函数来实现无监督方式训练．对于公开数据集LPBA40的3D脑磁共振图像配准任务，测试结果表明：本文所提出的方法与基线方法相比，具有较好的Dice得分、运行时间少且产生更好的微分同胚域，同时对噪声具有鲁棒性．

关键词: 深度学习, 群组配准, 变分推断, 可变形配准

Abstract:

To address the low precision of pairwise registration method based on the deep learning and the time-consuming nature of traditional registration algorithm, this paper presents a method of unsupervised end-to-end groupwise registration based on variational inference, as well as a registration framework based on normalized cross correlation (NCC) and prior knowledge. The framework can warp all images in the group into a common space and effectively control the deformation field of the regularization, and it doesn't need a real deformation field or a reference image. The estimation of deformation field by this method can be modeled as a probability generation model and solved by variational inference. Then unsupervised training is implemented with the help of spatial transformer network and loss function. The registration results of 3D brain magnetic resonance image from the public data set LPBA40 show that: compared with the baseline method, the proposed method has better Dice score, less running time, better diffeomorphisms domain, and is robust to noise.

Key words: deep learning, groupwise registration, variational inference, deformable registration

中图分类号:

O482.53

周勤, 王远军. 基于变分推断的磁共振图像群组配准[J]. 波谱学杂志, 2022, 39(3): 291-302.

Qin ZHOU, Yuan-jun WANG. Groupwise Registration for Magnetic Resonance Image Based on Variational Inference[J]. Chinese Journal of Magnetic Resonance, 2022, 39(3): 291-302.

图/表 11

图1

图2

图3

图4

图5

表1

图6

图7

图8

表2

图9

参考文献 26

1	WANG Y, JIANG F, LIU YReference-free brain template construction with population symmetric registration[J]. Med Biol Eng Comput, 2020,58 (9): 2083- 2093. doi: 10.1007/s11517-020-02226-5
2	MARTÍN-GONZÁLEZ E, SEVILLA T, REVILLA-ORODEA A, et alGroupwise non-rigid registration with deep learning: an affordable solution applied to 2D cardiac cine MRI reconstruction[J]. Entropy, 2020,22 (6): 687. doi: 10.3390/e22060687
3	GEE J C, REIVICH M, BILANIUK L, et alEvaluation of multiresolution elastic matching using mri data[J]. Proc Spie, 1991,1445,226- 234. doi: 10.1117/12.45220
4	RUECKERT D, SONODA L INonrigid registration using free-form deformations: application to breast mr images[J]. IEEE T Med Imaging, 1999,18 (8): 712- 721. doi: 10.1109/42.796284
5	THIRION J PImage matching as a diffusion process: an analogy with Maxwell's demons[J]. Med Image Anal, 2011,2 (3): 243- 260.
6	ZHONG J, PHUA D, QIU AQuantitative evaluation of lddmm, freesurfer, and caret for cortical surface mapping[J]. Neuroimage, 2010,52 (1): 131- 141. doi: 10.1016/j.neuroimage.2010.03.085
7	ASAMI T, BOUIX S, WHITFORD T J, et alLongitudinal loss of gray matter volume in patients with first-episode schizophrenia: dartel automated analysis and roi validation[J]. Neuroimage, 2012,59 (2): 986- 996. doi: 10.1016/j.neuroimage.2011.08.066
8	GUIMOND A, MEUNIER J, THIRION J PAverage brain models: a convergence study[J]. Comput Vis Image Und, 2000,77 (2): 192- 210. doi: 10.1006/cviu.1999.0815
9	SEGHERS D, D'AGOSTINO E, MAES F, et al. Construction of a brain template from mr images using state-of-the-art registration and segmentation techniques[C]// Medical Image Computing and Computer-Assisted Intervention--MICCAI 2004, 7th International Conference Saint-Malo, France, September 26-29, 2004, Proceedings, Part I. 2004.
10	WU G, JIA H, WANG Q, et alSharpmean: groupwise registration guided by sharp mean image and tree-based registration[J]. Neuroimage, 2011,56 (4): 1968- 1981. doi: 10.1016/j.neuroimage.2011.03.050
11	WU G, WANG Q, JIA H, et alFeature-based groupwise registration by hierarchical anatomical correspondence detection[J]. Hum Brain Mapp, 2012,33 (2): 253- 271. doi: 10.1002/hbm.21209
12	YANOVSKY I, THOMPSON P M, OSHER S, et al. Topology preserving log-unbiased nonlinear image registration: theory and implementation[C]// IEEE Conference on Computer Vision & Pattern Recognition. Minneapolis, Minnesota, USA: IEEE, 2007.
13	WANG Q, CHEN L Y YAP P T, et alGroupwise registration based on hierarchical image clustering and atlas synthesis[J]. Hum Brain Mapp, 2010,31,1128- 1140.
14	CHE T, ZHENG Y, CONG J, et alDeep group-wise registration for multi-spectural images from fundus images[J]. IEEE Access, 2019,7,27650- 27661. doi: 10.1109/ACCESS.2019.2901580
15	CHE T, ZHENG Y, SUI X, et al. Dgr-net: deep groupwise registration of multispectral images[C]// Information Processing in Medical Imaging - 26th International Conference , Hong Kong, china: IPMI, 2019: 706-717.
16	HAASE R, HELDMANN S, LELLMANN J. Deformable groupwise image registration using low-rank and sparse decomposition[EB/OL]. [2020-06-10].https://arxiv.org/abs/2001.03509.
17	DALCA A V, RAKIC M, GUTTAG J, et al. Learning conditional deformable templates with convolutional networks[C]//Neural Information Processing Systems 2019(NeurIPS 2019), Vancouver, BC, Canada, 2019: 804-816.
18	YU E M, DALCA A V, SABUNCU M R Learning conditional deformable shape templates for brain anatomy[M]. Lima: Machine Learning in Medical Imaging, 2020, 353- 352.
19	SIEBERT H, HEINRICH M P Deep groupwise registration of mri using deforming autoencoders[M]. Berlin: Springer, 2020, 236- 241.
20	HE Z Y, CHUNG A C S. Unsupervised end-to-end groupwise registration framework without generating templates[C]// 2020 IEEE International Conference on Image Processing (ICIP). IEEE, 2020: 375-379.
21	DALCA A V, BALAKRISHNAN G, GUTTAG J, et alUnsupervised learning of probabilistic diffeomorphic registration for images and surfaces[J]. Med Image Anal, 2019,57,226- 236. doi: 10.1016/j.media.2019.07.006
22	BALAKRISHNAN G, ZHAO A, SABUNCU M R, et alVoxelmorph: a learning framework for deformable medical image registration[J]. IEEE T Med Imaging, 2019,1788- 1800.
23	SHATTUCK D W, MIRZA M, ADISETIYO V, et alConstruction of a 3d probabilistic atlas of human cortical structures.[J]. Neuroimage, 2008,39 (3): 1064- 1080. doi: 10.1016/j.neuroimage.2007.09.031
24	王远军, 刘玉基于图像集拓扑中心的群体配准方法[J]. 波谱学杂志, 2018,35 (4): 60- 67.
	WANG Y J, LIU YGroup registration method based on topological center of images[J]. Chinese J Magn Reson, 2018,35 (4): 60- 67.
25	蔡文琴, 王远军基于磁共振成像的人脑图谱构建方法研究进展[J]. 波谱学杂志, 2020,37 (2): 241- 253.
	CAI W Q, WANG Y JAdvances in the construction of the human brain map based on magnetic resonance imaging[J]. Chinese J Magn Reson, 2020,37 (2): 241- 253.
26	刘可文, 刘紫龙, 汪香玉, 等基于级联卷积神经网络的前列腺磁共振图像分类[J]. 波谱学杂志, 2020,37 (2): 152- 161.
	LIU K W, LIU Z L, WANG X Y, et alProstate magnetic resonance image classification based on cascading convolutional neural networks[J]. Chinese J Magn Reso, 2020,37 (2): 152- 161.

方法	Avg. Dice^*	GPU运行时间/s^*	Percentages (%) of ${(\det ({J_\phi }(p)) \leqslant 0)^*}$	EM₁	EM₂
ANTs (Syn)	0.689 (0.036)		45.6 (19)	930.98	6.54
voxelmorph-diff	0.698 (0.030)	0.51 (0.03)	49.6 (11)	960.67	6.29
voxelmorph	0.672 (0.031)	0.49 (0.02)	48.9 (26)	969.46	7.08
本文方法	0.701 (0.025)	0.40 (0.02)	45.0 (18)	940.89	5.07

噪声方差（均值为0）	Avg. Dice^*	GPU运行时间/s^*	Percentages (%) of ${(\det ({J_\phi }(p)) \leqslant 0)^*}$	EM₁	EM₂
0.001	0.686 (0.026)	0.50 (0.02)	48.8 (21)	1330.97	89.54
0.002	0.683 (0.030)	0.60 (0.04)	49.3 (27)	1360.67	100.29
0.003	0.646 (0.031)	0.58 (0.02)	51.7 (19)	1469.46	99.08
0.004	0.656 (0.020)	0.62 (0.03)	54.5 (28)	1480.89	113.07

[1]	常晓,蔡昕,杨光,聂生东. 生成对抗网络在医学图像转换领域的应用[J]. 波谱学杂志, 2022, 39(3): 366-380.
[2]	王颖珊,邓奥琦,毛瑾玲,朱中旗,石洁,杨光,马伟伟,路青,汪红志. 基于3D VNetTrans的膝关节滑膜磁共振图像自动分割[J]. 波谱学杂志, 2022, 39(3): 303-315.
[3]	陈萌, 耿辰, 李郁欣, 耿道颖, 鲍奕仿, 戴亚康. 基于模糊标签和深度学习的TOF-MRA影像脑动脉瘤自动检测[J]. 波谱学杂志, 2022, 39(3): 267-277.
[4]	王振宇, 王颖珊, 毛瑾玲, 马伟伟, 路青, 石洁, 汪红志. 基于Dense-UNet++的关节滑膜磁共振图像分割[J]. 波谱学杂志, 2022, 39(2): 208-219.
[5]	霍璐,胡晓欣,肖勤,顾雅佳,褚旭,姜娈. 基于nnU-Net的乳腺DCE-MR图像中乳房和腺体自动分割[J]. 波谱学杂志, 2021, 38(3): 367-380.
[6]	刘鹏, 钟玉敏, 王丽嘉. 基于密集多尺度U-net网络的电影心脏磁共振图像右心室自动分割[J]. 波谱学杂志, 2020, 37(4): 456-468.
[7]	赵尚义, 王远军. 基于磁共振图像和改进的UNet++模型区分阿尔茨海默症患者和健康人群[J]. 波谱学杂志, 2020, 37(3): 321-331.
[8]	宫进昌, 王宇, 王远军. 结合小波融合和深度学习的脑胶质瘤自动分割[J]. 波谱学杂志, 2020, 37(2): 131-143.
[9]	程慧涛, 王珊珊, 柯子文, 贾森, 程静, 丘志浪, 郑海荣, 梁栋. 基于深度递归级联卷积神经网络的并行磁共振成像方法[J]. 波谱学杂志, 2019, 36(4): 437-445.
[10]	苏新宇, 王丽嘉, 聂生东, 胡立伟, 钟玉敏. 基于心脏磁共振短轴电影图像的右心室分割新进展[J]. 波谱学杂志, 2019, 36(3): 377-391.