Whole heart segmentation using image registration
Image registration and atlas-based segmentation:
Image registration can be considered as a process to compute one or one series of spatial transformations which can map one or one series of medical images to other one with accurate spatial correspondence. The result of an image registration process can be the transformations, or the transformed resultant images, or both of them.
Image registration provides the capability to fuse, compare, or propagate the information from different images in the same coordinate system.
Image registration provides an alternative solution for segmentation propagation [TMI10]. Registration-based atlas propagation and segmentation works by registering a pre-constructed atlas, having every region of interested segmented, to unseen images to achieve fully automated segmentation. With a successful registration between the atlas and unseen image, the segmentation of the unseen image becomes available. My interest includes developing registration techniques to overcome the problems in segmentation propagation, such as the robustness against low quality images, the flexibility of the deformation model which determines the segmentation accuracy, and the diffeomorphism.
Whole heart segmentation error map
Whole heart segmentation
JHE13, Review Article]:
In contrast with ventricle or myocardium segmentation, whole heart segmentation provides the volume or delineation of ventricles, myocardium, atria, and sometimes great vessels as well if they are of interests. To achieve a fully automated process, we developed an atlas propagation based method using image registration techniques. This method employs the two newly developed registration algorithms: locally affine registration method (LARM) and the free-form deformation registration with adaptive control point status
We validated this segmentation framework in cardiac MRI involving thirty-seven subjects and nine types of pathologies. The MRI tool has been exported to two research groups in University of Oxford and St. Thomas Hospital, King's College London.
The segmentation framework also demonstrated potential of being applicable to cardiac ultrasound using local phase registration and compounding techniques [isbi10].
The results of cardiac segmentation is essential for studying changes of cardiac functions. It can enable cardiologists to more accurately detect early symptoms of cardiac diseases in clinical applications, and/ or better understand relations between cardiac outward features and special diseases/ treatments in clinical researches.
Spatially encoded mutual information (SEMI/SIEMI) [ipmi09,wbir10,TMI11]: |
Traditional mutual information may be inappropriate in some applications when the intensity class correspondence between the images are not bijective, meaning the mapping of tissue type related intensity classes is not one-to-one. The non-bijective mapping can consequently bring ambiguity or noise into the tissue correspondence. This effect brings registration errors, particularly in nonrigid registration. SIEMI includes the spatial information to tackle this problem and has demonstrated promising results in brain MRI, cardiac MRI, and contrast enhanced MRI of the liver.
Left figure: T1-weighted brain image without intensity non-uniformity (INU) filed, the resultant deformation field of registering the image without INU to an image with INU using normalized mutual information measure, and the resultant deformation field of the registering using SIEMI registration. The bar on right indicates the scale of the magnitude of the two deformation fields.
Locally affine registration method (LARM) and
Dynamic Resampling And distance Weighting interpolation (DRAW)
Locally affine transformation is an attractive registration alternative for some applications where a single global affine transformation cannot provide enough accuracy while a non-rigid registration would affect incorrectly the local topology. LARM assigns a local affine transformation to each substructure for locally affine registration, while globally it is a deformable registration. LARM can be used to further initialize the substructures within the images after a global affine registration. This initialization is crucial for the following nonrigid registration to maintain the local topology, and then achieve good robustness in inter-subject cases.
Many nonrigid registration algorithms themselves do not provide the inverse of the resultant transformation. This inverse transformation can be crucial in some applications. For example, LARM requires a pre-definition of the local regions in the reference image. However, in the segmentation propagation framework, the prior definition is only available from the floating image, the atlas. In this case, it is required to compute an inverse transformation from the result of the LARM process. DRAW is such a method we designed for computing the inverse of bijective deformation fields.
|Invited talks:||2011-06-21 A Registration-Based Atlas Propagation Framework for Automatic Whole Heart Segmentation (ppt), Fields-MITACS Conference on Mathematics of Medical Imaging|
|2011-06-15 Extracting prior shape information without a training stage for image segmentation and registration, International Society for Computer Assisted Orthopaedic Surgery, workshop Statistical Shape Modelling and its Applications in CAOS|
|2011-03-21 Cardiac segmentation and functional analysis, GE Healthcare|
|Committee member:||International Conference on Functional Imaging and Modeling of the Heart (FIMH)|
|Review for:|| IEEE Transactions on Medical Imaging |
Medical Image Analysis
Computer Vision and Image Understanding
Computerized Medical Imaging and Graphics
Image and Vision Computing
Computer Methods and Programs in Biomedicine
Journal Computing and Informatics
International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI)
International Conference on Functional Imaging and Modeling of the Heart (FIMH)
International Workshop on Medical Imaging and Augmented Reality (MIAR)
Challenges and Methodologies of Fully Automatic Whole Heart Segmentation: A Review.
Journal of Healthcare Engineering 4 (3): 371–407, 2013
W Shi, M Jantsch, P Aljabar, L Pizarro, W Bai, H Wang, D O’Regan, X Zhuang* and D Rueckert*: Temporal sparse free-form deformations. Medical Image Analysis 17 (7): 779–789, 2013 link
L. Liu, W. Shi, D. Rueckert, M. Hu, S. Ourselin, X. Zhuang*: Model-Guided Directional Minimal Path for Fully Automatic Extraction of Coronary Centerlines from Cardiac CTA. Medical Image Computing and Computer-Assisted Intervention (MICCAI) 2013
W. Shi, J. Caballero, C. Ledig, X. Zhuang, W. Bai, K. Bhatia, A. Marvao, T. Dawes, D. O'Regan, D. Rueckert: Cardiac image super-resolution with global correspondence using multi-atlas PatchMatch. Medical Image Computing and Computer-Assisted Intervention (MICCAI) 2013
J Xi, P Lamata, S Niederer, S Land, W Shi, X Zhuang, et al.: The estimation of patient-specific cardiac diastolic functions from clinical measurements Medical image analysis 17 (2), 133–146 2013
C Tobon-Gomez, et al.: Benchmarking framework for myocardial tracking and deformation algorithms: An open access database. Medical image analysis 17 (6), 632–648 2013
X Zhuang, et al.: A registration and atlas propagation based framework for automatic whole heart segmentation of CT volumes SPIE Medical Imaging, 86693W-86693W-8 2013
X Zhuang, W Shi, H Wang, D Rueckert, S Ourselin Computation on shape manifold for atlas generation: application to whole heart segmentation of cardiac MRI SPIE Medical Imaging, 866941-866941-7 1 2013
H Wang, W Shi, X Zhuang, X Wu, KP Tung, S Ourselln, P Edwards, D Rueckert Landmark detection and coupled patch registration for cardiac motion tracking SPIE Medical Imaging, 86690J-86690J-6 2013
H Huang, X Zhuang, et al.: Computer-aided scheme for functional index computation of left ventricle in cardiac CTA: segmentation and partitioning of left ventricle SPIE Medical Imaging, 867017-867017-6 2013
| Shi, W., Zhuang, X., Luong, D., Wang, H., Tung, K., Duckett, S., Edwards, P., Razavi, R., Ourselin, S., Rueckert, D.: A Comprehensive Cardiac Function Analysis Framework Using Both Untagged And 3D Tagged MR Images.
IEEE Transactions on Medical Imaging, 31(6), 1263 – 1275, 2012.
W Shi, X Zhuang, L Pizarro, W Bai, H Wang, KP Tung, P Edwards, D Rueckert: Registration using sparse free-form deformations. Medical Image Computing and Computer-Assisted Intervention–MICCAI 2012, 659-666, 1, 2012 (MICCAI Young Scientist Award Honorable Mention)
SG Duckett, W Shi, X Zhuang, et al.: Cardiac MRI: understanding myocardial motion to predict remodelling pre cardiac resynchronisation therapy Heart 98 (Suppl 1), A6-A7 1 2012
H Wang, W Shi, X Zhuang, et al.: Automatic Cardiac Motion Tracking Using Both Untagged and 3D Tagged MR Images Statistical Atlases and Computational Models of the Heart. Imaging and Modelling Challenges 2012
W Shi, X Zhuang et al.: A multi-image graph cut approach for cardiac image segmentation and uncertainty estimation Statistical Atlases and Computational Models of the Heart. Imaging and Modelling Challenges 2012
|Zhuang, X., Arridge, S., Hawkes, D. J., Ourselin, S.: A Nonrigid Registration Framework Using Spatially Encoded Mutual Information and Free-Form Deformations. IEEE Transactions on Medical Imaging, 30(10), 1819-1828, 2011.   link pdf|
|Zhuang, X., Rhode, K., Razavi, R., Hawkes, D. J., Ourselin, S.: A Registration-Based Propagation Framework for Automatic Whole Heart Segmentation of Cardiac MRI. IEEE Transactions on Medical Imaging, 29 (9), 1612-1625, 2010.   link pdf|
|Zhuang, X., Shi, W., Duckette, S., Wang, H., Razavi, R., Hawkes, D., Rueckert, D., Ourselin, S.: A framework combining multi-sequence MRI for fully automated quantitative analysis of cardiac global and regional functions. Oral presentation in Functional Imaging and Modeling of the Heart (FIMH), LNCS 6666, pp. 367--374, 2011.   link|
|Shi, W., Zhuang, X., Wang, H., Duckett, S., Oregan, D., Edwards, P., Ourselin, S., Rueckert, D.: Automatic segmentation from cardiac cine MRI with different pathologies using registration and multiple component EM estimation. Functional Imaging and Modeling of the Hear (FIMH), LNCS 6666, pp. 163--170, 2011. link|
|Zhang, D., Zhuang, X., Ourselin, S., Rueckert, D.: Motion tracking of left ventricle and coronary arteries in 4D CTA. SPIE Medical Imaging 2011. pdf|
|Shi, W., Zhuang, X., Wolz, R., Duckett, S., Tung, K-P, Wang, H, Ourselin, S., Edwards, P., Razavi, R., Rueckert, D.: A multi-image graph cut approach for cardiac image segmentation and uncertainty estimation, MICCAI Statistical Atlases and Computational Models of the Heart Workshop (2011) pdf|
|Wang, H., Shi, W., Zhuang, X., Duckett, S., Tung, K-P, Edwards, P., Razavi, R., Ourselin, S., and Rueckert, D.: Automatic cardiac motion tracking using both untagged and 3D tagged MR images, MICCAI Statistical Atlases and Computational Models of the Heart Workshop (2011) pdf|
|Zhuang, X., Leung, K., Rhode, K., Razavi, R., Hawkes, D. J., Ourselin, S. (2010): Whole Heart Segmentation of Cardiac MRI Using Multiple Path Propagation Strategy. Medical Image Computing and Computer Assisted Intervention (MICCAI'10), LNCS 6361, 435-443, 2010. DOI: 10.1007/978-3-642-15705-9_53   link|
|Zhuang, X., Hawkes, D. J., Ourselin, S. (2010). Spatial Information Encoded Mutual Information for Nonrigid Registration. Oral presentation at International Workshop on Biomedical Image Registration (WBIR '10), LNCS 6204, 246-257, 2010.   pdf|
|Zhuang, X., Yao, C., Ma, Y. L., Hawkes, D. J., Penney, G., Ourselin, S. (2010). Registration-Based Propagation for Whole Heart Segmentation from Compounded 3D Echocardiography. Oral presentation at IEEE International Symposium on Biomedical Imaging (ISBI '10), 1093-1096, 2010.   link|
|Zhuang, X., Hawkes, D. J., Ourselin, S. (2009). Unifying encoding of spatial information in mutual information for nonrigid registration. In: J.L. Prince, D.L. Pham, and K.J. Myers (Eds.), 21st biennial International Conference on Information Processing in Medical Imaging (IPMI '09), Lecture Notes in Computer Science (LNCS) 5636, 491-502, 2009.   link|
|Zhuang, X., Rhode, K., Razavi, R., Hawkes, D. J., Ourselin, S. (2009). Free-Form Deformations Using Adaptive Control Point Status for Whole Heart MR Segmentation. In: Functional Imaging and Modeling of the Heart (FIMH '09), Lecture Notes in Computer Science (LNCS) 5528, 303-311. 2009.   link|
| Zhuang, X., Rhode, K., Arridge, S., Razavi, R., Hill, D., Hawkes, D. J., Ourselin, S. (2008). An atlas-based segmentation propagation framework using locally affine registration – Application to automatic whole heart segmentation. Oral presentation at Medical Image Computing and Computer Assisted Intervention (MICCAI '08), Lecture Notes in Computer Science (LNCS) 5242, 425-433, 2008.
(runner-up for MICCAI Young Scientist Award, New York University)
|Zhuang, X., Ourselin, S., Razavi, R., Hill, D. L. G, Hawkes, D. J. (2008). Automatic Whole Heart Segmentation Based on Atlas Propagation with a Priori Anatomical Information. In: Medical Image Understanding and Analysis (MIUA) 2008, 29-33, 2008. pdf|
|Zhuang, X., Hawkes, D. J., Crum, W. R., Boubertakh, R., Uribe, S., Atkinson, D., Batchelor, P., Schaeffter, T., Razavi, R., Hill ,D. L. G. (2008). Robust Registration between Cardiac MRI Images and Atlas for Segmentation Propagation. In: SPIE Vol. 6914 Medical Imaging 2008: Image Processing, 6914, 07, 2008. pdf|
|Zhuang, X., Gu, L. (2006). Normal Vector Information Registration and Comparisons with Mutual Information. In: IEEE Engineering in Medicine and Biology Society. 1:3827-30, 2006.   pdf|
|Xu, J., Gu, L., Zhuang, X., Peters, T.~M. (2005). A Novel Multi-stage 3D Medical Image Segmentation: Methodology and Validation. In: Hao. Y., etc. (eds.): Computational Intelligence and Security. Lecture Notes in Artificial Intelligence (LNAI), Vol. 3801 (2005) 884-889, 2005. link|
|Zhuang, X., Gu, L., Xu, J. (2005). Medical Image Alignment by Normal Vector Information. In: Hao. Y., etc. (eds.): Computational Intelligence and Security. Lecture Notes in Artificial Intelligence (LNAI), Vol. 3801 (2005) 890-895, 2005. link|