|
|
|
| Lung 4D-CT Image Registration Based on Regression Prediction |
| Liu Yueliang1,2, Fang Shiting1,2, Zhang Yu1,2* |
1School of Biomedical Engineering, Southern Medical University, Guangzhou 510515, China
2Guangdong Provincial Key Laboratory of Medical Image Processing, Southern Medical University, Guangzhou 510515, China |
|
|
|
|
Abstract To improve the low accuracy of lung 4D-CT image registration due to intensity inhomogeneity and large local deformation, a regression-based method for predicting initial deformation is presented. The core idea of this method is that the image information of different phase corresponding to the floating image is used to predict initialize deformation field. Firstly, the common registration algorithmis used to register the different phase images to the reference image to get the corresponding deformation field. Then, the image and the corresponding deformation field are divided into patches to build training set. Multi-dimensional support vector regressor is used to learn a regression model with respect to the correlation between patch appearances and their respective deformations. The floating image is then input into the regression model to achieve the initial deformation field prediction, which is used to construct intermediate image. Finally, the registration of intermediate image and reference image are refined. The experimental results show that the proposed method can effectively register lung 4D-CT images. A public dataset provided by the DIR-lab at the University of Texas M.D. Anderson Cancer Center (Houston, TX) was employed to evaluate the proposed method. Quantitative evaluation results indicate that the proposed method significantly decreases SSD (18.97±5.75,P<0.05) compared with Active Demons algorithm(49.34±23.92) and Spectral Log-Demons algorithm (31.81±15.09),significantly increases CC (0.980 ± 0.006,P<0.05) compared with Active Demons algorithm(0.952±0.022) and Spectral Log-Demons algorithm (0.967±0.015). Qualitative evaluation results also show that our approach can yield superior registration performance.
|
|
Received: 09 December 2016
|
|
|
|
|
|
[1] Sluimer I, Prokop M, Van GB. Toward automated segmentation of the pathological lung in CT [J]. IEEE Transactions on Medical Imaging, 2005, 24(8):1025-1038.
[2] Sarrut D, Boldea V, Miguet S, et al. Simulation of four-dimensional CT images from deformable registration between inhale and exhale breath-hoet al.ld CT scans [J]. Medical Physics, 2006, 33(3):605-617.
[3] Yin Y, Hoffman EA, Lin CL, et al. Mass preserving nonrigid registration of CT lung images using cubic B-spline [J]. Medical Physics, 2009, 36(9):4213-4222.
[4] Gorbunova V, Sporring J, Lo p, Loeve M, et al. Mass preserving image registration for lung CT [J]. Medical Image Analysis,2012,16(4):786-795
[5] Castillo E, Castillo R, Fuentes D, et al.Computing global minimizers to a constrained B-spline image registration problem from optimal L1 perturbations to block match data [J]. Medical Physics, 2014, 41(4):041904-041904.
[6] Yang J, Wright J, Huang TS, et al. Image super-resolution via sparse representation [J].IEEE Transactions on Image Processing, 2010,19(11):2861-2873.
[7] Wang Z, Zhen X, Tay KY, et al. Regression segmentation for M3 spinal images [J]. IEEE Transactions on Medical Imaging, 2014, 34(8):1640-1648.
[8] Lombaert H, Grady L, Pennec X, et al.Spectral Log-Demons: Diffeomorphic Image Registration with Very Large Defoemation [J]. International Journal of Computer Vision, 2014, 107(3):254-271.
[9] Kim M, Wu G, Wang Q, et al. Improved image registration by sparse patch-based deformation estimation [J]. NeuroImage, 2015, 105:257-268.
[10] Wang T, Cao L, Yang W, et al. Adaptive patch-based POCS approach for super resolution reconstruction of 4D-CT lung data [J].Physics in Medicine and Biology, 2015, 60(15):5939-5954.
[11] Rez-Cruz F, Camps-Valls G, Soria-Olivas E,et al. Multi-dimensional function approximation and regression estimation [C] //International Conference on Artificial Neural Networks. Madrid: CANN, 2002: 757-762.
[12] Sanchez-Fernandez M, De-Prado-Cumplido M, Arenas-Garcia J, et al. SVM multiregression for nonlinear channel estimation in multiple-input multiple-output systems [J]. IEEE Transactions on Signal Processing, 2004, 52(8):2298-2307.
[13] 张学工.模式识别[M].3版.北京:清华大学出版社,2010:109,114-115.
[14] Scholkopf B, Somla A, Learning with Kernels [M]Cambridges: MIT Press, 2002.
[15] Perez-Cruz F, Alarcon-Diana P, Navia-Vazquez A, et al.Fast training of support vector classifiers [J]. Advances in Neural Information Processing Systems, 2001: 1849-1870.
[16] Castillo E, Castillo R, Guerra R, et al.A framework for evaluation of deformable image registration spatial accuracy using large landmark point sets [J]. Physics in Medicine and Biology, 2009, 54(7):1849-1870.
[17] Schmidt-Richberg A, Werner R, Handels H, et al.Estimation of slipping organ motion by registration with direction-dependent regularization [J]. Medical Image Analysis, 2012, 16(1):150-159. |
|
|
|
