3D vasculature segmentation using localized hybrid level-set method
© Hong et al.; licensee BioMed Central. 2014
Received: 27 July 2014
Accepted: 27 November 2014
Published: 16 December 2014
Intensity inhomogeneity occurs in many medical images, especially in vessel images. Overcoming the difficulty due to image inhomogeneity is crucial for the segmentation of vessel image.
This paper proposes a localized hybrid level-set method for the segmentation of 3D vessel image. The proposed method integrates both local region information and boundary information for vessel segmentation, which is essential for the accurate extraction of tiny vessel structures. The local intensity information is firstly embedded into a region-based contour model, and then incorporated into the level-set formulation of the geodesic active contour model. Compared with the preset global threshold based method, the use of automatically calculated local thresholds enables the extraction of the local image information, which is essential for the segmentation of vessel images.
Experiments carried out on the segmentation of 3D vessel images demonstrate the strengths of using locally specified dynamic thresholds in our level-set method. Furthermore, both qualitative comparison and quantitative validations have been performed to evaluate the effectiveness of our proposed model.
Experimental results and validations demonstrate that our proposed model can achieve more promising segmentation results than the original hybrid method does.
The reconstruction of blood vessel plays an important role for many clinical tasks such as diagnosis of vessel diseases, surgery planning and blood flow simulation . The first task to vessel reconstruction is to identify the vessel region from initial vessel images. This is usually achieved by using the technique of segmentation, which plays an important role in medical image processing. Although many segmentation techniques have been proposed for various image segmentation problems, blood vessel image segmentation still remains a challenging task , due to the high complexity of vessel branching and thinning geometry as well as the decrease in image contrast from the root of the vessel to its thin branches .
Among all the segmentation techniques, the deformable model (i.e. active contour)  is one of the most popular approaches during the past two decades. The basic idea of deformable model method is to evolve the initial curve/surface to the boundaries of target objects driven by the combination of internal forces determined by the geometry of the evolving curve and the external forces induced from the image . Depending on the way the underlying curved contour is represented, the deformable active contour can be implemented either parametrically or implicitly . In parametric models, active contours are explicitly represented in parametric forms in a Lagrangian framework . A multitude of powerful methods based on parametrical deformable model have been proposed for the segmentation of medical image . Klein et al.  presented a method for the extraction of vessel boundaries using deformable surface models represented by B-spline functions. However, the parameterized deformable models are difficult to handle the complex topological changes for the whole vasculatures during evolution, as reparameterization is a tough problem to be solved for parametric model . On the other hand, active contours can be implicitly represented as level-set  function, which is embedded in higher dimensional spaces and evolves according to the Partial Differential Equation (PDE) in an Eulerian framework . Compared with parametric model, level-set model is easier to perform shape operations, and adapt freely into complex topologies of objects. Therefore, various level-set based active models have been developed and applied to image segmentation. Some of the well-known models are the geodesic active contour model  that utilizes image gradient to stop evolving contours on the object boundaries, and Chan-Vese model  that solves the leakage problem by using region information of the target boundary for segmentation. There are also several localized active models [13, 14] introducing the local image fitting energy to extract the local image information been presented to improve the performance of the active contours. Furthermore, some techniques combining edge and region information [4, 15, 16] have been developed and applied to segmentation of medical imaging.
In the area of 3D vessel image segmentation, Lorigo et al.  proposed the ’CURVES’ algorithm to extract vessels by using a geodesic active contour based on the codimension two level set method . Nain et al.  proposed a region-based geometric deformable model for MRA segmentation, which is driven by the combination of image statistics and shape information. Yan and Kassim  presented a segmentation scheme for the extraction of vasculature from MRA images by using capillary active contour. In , the vessel contours are segmented from the enhanced dataset by means of a level set evolution, which relies on the image intensity statistics to either expand or shrink the evolving contour.In , a shape functional is used to regularize the ‘CURVES’ and flux maximizing flow functional , which is claimed to be a promising tool to improve the efficiency of both techniques. Law and Chung  presented a vessel extraction approach based on a weighted local variance based edge detection scheme, which is effective for the segmentation of vessel image with low contrast edges. Recently, Wang and Jiang presented a nonparametric shape constrained algorithm for segmentation of coronary arteries in computed tomography images within the framework of active contours . However, as most of these active contour models only take the edge information or region information of the image into account, they may have difficulty in correctly extracting tiny vessels from images, due to the low image contrast of the thin vessel branches.
In this paper, a localized hybrid level-set method for the segmentation of 3D vessel image is proposed. Our technique is inspired by Zhang et al.’s hybrid level-set model . The hybrid model integrates both region and boundary information for the segmentation of medical images, in which the boundary information can help to detect the precise location of the target object and the region information can help to prevent the boundary leakage problem. However, the original hybrid model utilized a preset global threshold indicating the lower bound of target object to specify the term concerning regional information, which is not quite appropriate, especially for medical images with intensity inhomogeneity. In fact, intensity inhomogeneity occurs in many medical images , especially in vessel images, as the vessel branching is highly complex and the image contrast decreases from the root of the vessel to its thin branches . On the other hand, our proposed localized hybrid level-set model utilizes the automatically calculated dynamic thresholds to indicate the lower bound of target object to specify the term concerning regional information instead of using a preset global threshold, which is crucial for the segmentation of vessel images . In our proposed technique, the local intensity information is firstly embedded into a region-based contour model, and then incorporated into the level-set formulation of the geodesic active contour model. Compared with the global threshold based method, the use of locally specified dynamic thresholds enables the extraction of the local image information, which is essential for the segmentation of vessel images. Experimental results on 3D medical datasets and validations are presented to demonstrate the strengths of our localized hybrid level-setmethod.
where k∈[0.5,1] is an adjustment coefficient, and K σ is the Gaussian kernel function with a localization property, such as with a scale parameter σ. The coefficient k is effective for preventing the active contours stopping evolving inside the target areas before reaching the boundary, which makes our proposed model more robust for the segmentation of 3D medical dataset. The first term of functional (1) defines the region term in our localized hybrid level-set model. The second term is equal to the geodesic active contour functional, which is the same as the second term of the original hybrid model. In addition to keeping the advantages of original hybrid model, the localized hybrid model is able to extract more accurate local image information by utilizing the locally specified dynamic thresholds μ(u) to indicate the lower bound of target object, which is essential for segmenting images with intensity inhomogeneity.
The original Heaviside function is discontinuous and is unable to achieve smooth transition at the boundary. In practice, this problem is generally solved using a kind of smooth Heaviside function, which is usually called smooth step function or smooth unit step function. Depending on the problem of applications, different smooth step functions have been introduced to approximate smoothly the Heaviside function . More recently, a piecewise polynomial smooth unit step functions is developed and used for geometric shape design, which can be considered as a generalization of the Heaviside unit step function .
Reinitialize ϕ (k) to be SDF;
Calculate the localized thresholds μ(u) automatically according to Equation 3;
Compute the local region energy functional δ ε (ϕ (k))(I−μ(u));
Update ϕ (k) to ϕ (k) ′ using ϕ (k) ′=ϕ (k)+Δ t α δ ε (ϕ (k))(I−μ(u));
Reinitialize ϕ (k) ′ to be SDF;
Update ϕ (k) ′ to obtain ϕ (k+1) via ϕ (k+1)=ϕ (k) ′+Δ t β δ ε (ϕ (k))d i v(g(|∇I|)∇ϕ (k)).
Results and discussion
Results of our localized hybrid model
The proposed localized hybrid model has been applied to more than ten 3D medical datasets for the segmentation of vasculatures. The datasets are supplied by Intelligent Bioinformatics Systems Division, Institute of Automation, the Chinese Academy of Sciences, in the format of DICOM (Digital Imaging and Communications in Medicine). In all these experiments, the same parameters are used, namely, Δ t=4.0, α=0.01, β=0.5, σ=3.0, ε=1.0, and k=0.9. In the following, we present some typical segmentation results as well as the Maximum Intensity Projections (MIPs) of the corresponding original datasets. MIPs are very useful to evaluate current 3D angiography images since the overall shapes and paths of the vessel can be clearly visualized .
Comparison with the original hybrid level-set model
In this section, we investigate why the original hybrid model may fail in the segmentation of real medical data with intensity inhomogeneity whereas our localized model works. According to our prior knowledge, the intensity (i.e. gray level) of most vessels in the first dataset (i.e. cerebral MRA images) is larger than 200. However, the intensity of some small vessels and their branches can be varied, for some of them can even be lower than 100. Therefore, for the original model, it is quite difficult to predefine the global threshold μ indicating the lower bound of gray level of target vessel. If the global threshold μ were set to be 200, it would be unable to extract the small vasculatures with intensity smaller than 200; if μ were set to be 100, the segmentation result could include some irrelevant objects, which would lead to inaccurate extraction of vasculatures.
In a word, this qualitative comparison indicates that our localized hybrid model can achieve a more satisfactory segmentation result of the vascular structures than the original model with different μ s.
where A and B are the target and obtained segmentation sets, and n is the voxel number. The smaller value of SE means the more accuracy of the segmentation result.
Segmentation Errors of Chan-Vese (CV) model, Region-Scalable Fitting Energy (RSFE) based method, Nonparametric Shape Prior Constrained Active Contour (NSPCAC) model, Original Hybrid Model and Localized Hybrid Model for different Gaussian noise variance levels applied on the test dataset
RSFE based method
Original Hybrid Model
Localized Hybrid Model
Vessel image segmentation is an essential step for many clinical tasks such as diagnosis of vessel diseases, surgery planning and blood flow simulation. However, due to the high complexity of vessel branching and thinning geometry, blood vessel image segmentation still remains a challenging task. In this paper, a localized hybrid level-set model for vessel image segmentation is proposed. Our technique is inspired by Zhang et al.’s hybrid level-set model . However, instead of using a preset global threshold to specify the term concerning regional information, the proposed method compute the local thresholds automatically to indicate the lower bound of target object, which is crucial for the segmentation of vessel images. Compared with the global threshold based method, the use of locally specified dynamic thresholds enables the extraction of the local image information. Our proposed method has been successfully applied to over ten 3D medical datasets for the segmentation of vasculatures. Some experimental results are presented to demonstrate the strengths of our localized hybrid level-set method. In addition, both the qualitative and the quantitative validations have been performed to indicate that our proposed model can achieve more accurate segmentation results than original hybrid model and CV model do.
This work is supported by the Fundamental Research Funds for the Central Universities (Grant No. 2013121030), grant of Scientific and Technology Emphasis Project of Fujian Province (2011H0031), projects from National Natural Science Foundation of China (No.61100106, No.61174161, No.61402389), and MOE (Ministry of Education in China) Project of Humanities and Social Sciences (Grant No. 11YJC870027). The authors would like to thank the reviewers for their constructive comments.
- Nain D, Yezzi A, Turk G: Vessel segmentation using a shape driven flow. Lect Notes Comput Sci 2004, 3216: 51–59. 10.1007/978-3-540-30135-6_7View ArticleGoogle Scholar
- Lesage D, Angelini ED, Bloch I, Funka-Lea G: A review of 3d vessel lumen segmentation techniques: Models, features and extraction schemes. Med Image Anal 2009, 13: 819–845. 10.1016/j.media.2009.07.011View ArticleGoogle Scholar
- Kass M, Witkin A, Terzopoulos D: Snakes: active contour models. Int J Comput Vis 1988, 1: 321–331. 10.1007/BF00133570View ArticleGoogle Scholar
- Zhang Y, Matuszewski BJ, Shark LK, Moore CJ: Medical image segmentation using new hybrid level-set method. In Proceedings of Fifth International Conference BioMedical Visualization: Information Visualization in Medical and Biomedical Informatics. London, UK: IEEE 2008, 71–76.Google Scholar
- A Review on the Current Segmentation Algorithms for Medical Images http://www.researchgate.net/publication/221116197_A_Review_on_the_Current_Segmentation_Algorithms_for_Medical_Images
- Li C, Xu C, Gui C, Fox MD: Level set evolution without re-initialization: a new variational formulation. In Proceedings of IEEE International Conference on Computer Vision and Pattern Recognition (CVPR). San Diego. USA: IEEE 2005, 430–436.Google Scholar
- McInerney T, Terzopoulos D: Deformable models in medical image analysis: a survey. Med Image Anal 1996,1(2):91–108. 10.1016/S1361-8415(96)80007-7View ArticleGoogle Scholar
- Klein AK, Lee F, Amini AA: Quantitative coronary angiog-raphy with deformable spline models. IEEE Trans Med Imaging 1997, 16: 468–482. 10.1109/42.640737View ArticleGoogle Scholar
- Antiga L: Patient-specific modelling of geometry and blood flow in large arteries. PhD thesis. 2002.Google Scholar
- Osher S, Sethian JA: Fronts propagating with curvature dependent speed: algorithms based on hamilton-jacobi formulations. J Comput Phys 1988, 79: 12–49. 10.1016/0021-9991(88)90002-2MathSciNetView ArticleGoogle Scholar
- Caselles V, Kimmel R, Sapiro G: Geodesic active contours. Int J Comput Vis 1997, 22: 61–79. 10.1023/A:1007979827043View ArticleGoogle Scholar
- Chan T, Vese L: Active contours without edges. IEEE Trans Image Processs 2001,10(2):266–277. 10.1109/83.902291View ArticleGoogle Scholar
- Li C, Kao C, Gore J, Ding Z: Implicit active contours driven by local binary fitting energy. In Proceedings of 2007 IEEE Conf. Computer Vision and Pattern Recognition. Minneapolis, USA: IEEE 2007, 1–7.Google Scholar
- Li C, Kao C-Y, Gore JC, Ding Z: Minimization of Region-Scalable Fitting Energy for Image Segmentation. IEEE Trans Image Process 2008,17(10):1940–1949.MathSciNetView ArticleGoogle Scholar
- Hong Q, Wang B: Segmentation of vessel images using a local hybrid level-set method. In Proceedings of 6th International Congress on Image and Signal Processing. Hangzhou, China: IEEE 2013, 631–635.Google Scholar
- Li B, Chui C, Chang S, Ong S: A new unified level set method for semi-automatic liver tumor segmentation on contrast-enhanced ct images. Expert Syst Appl 2012, 39: 9661–9668. 10.1016/j.eswa.2012.02.095View ArticleGoogle Scholar
- Lorigo LM, Faugeras O, Grimson WEL, Keriven R, Kikinis R, Nabavi A, Westin C-F: Curves: Curve evolution for vessel segmentation. Med Image Anal 2001, 5: 195–206. 10.1016/S1361-8415(01)00040-8View ArticleGoogle Scholar
- Ambrosio L, Soner H. M: Level set approach to mean curvature flow in arbitrary codimension. J Differential Geometry 1996, 43: 693–737.MathSciNetGoogle Scholar
- Yan P, Kassim AA: Segmentation of volumetric mra images by using capillary active contour. Med Image Anal 2006,10(3):317–329. 10.1016/j.media.2005.12.002View ArticleGoogle Scholar
- Wu X, Luboz V, Krissian K, Cotin S, Dawson S: Segmentation and reconstruction of vascular structures for 3d real-time simulation. Med Image Anal 2010,15(1):22–34.View ArticleGoogle Scholar
- Gooya A, Liao H, Matsumiya K, Masamune K, Masutani Y, Dohi T: A variational method for geometric regularization of vascular segmentation in medical images. IEEE Trans Image Process 2008,17(8):1295–1312.MathSciNetView ArticleGoogle Scholar
- Vasilevskiy A, Siddiqi K: Flux maximizing geometricflows. IEEE Trans Pattern Anal Mach Intell 2002,24(12):1565–1578. 10.1109/TPAMI.2002.1114849View ArticleGoogle Scholar
- Law MWK, Chung ACS: Weighted local variance-based edge detection and its application to vascular segmentation in magnetic resonance angiography. IEEE Trans Med Imaging 2007,26(9):1224–1241.View ArticleGoogle Scholar
- Wang Y, Jiang H: A nonparametric shape prior constrained active contour model for segmentation of coronaries in cta images. Comput Math Methods Med 2014, 2014: 1–11.Google Scholar
- Li Q, Griffiths JG, Ward J: Constructive implicit fitting. Comput Aided Geom Des 2006,23(1):17–44. 10.1016/j.cagd.2005.04.011MathSciNetView ArticleGoogle Scholar
- Li Q, Tian J: Partial shape-preserving splines. Comput Aided Des 2011, 43: 394–409. 10.1016/j.cad.2011.01.007View ArticleGoogle Scholar
- Osher S, Fedkiw R: Level Set Methods and Dynamic Implicit Surfaces. New York: Springer; 2002.Google Scholar
- Orkisz MM, Bresson C, Magnin IE, Champin O, Douek PC: Improved vessel visualization in mr angiography by nonlinear anisotropic filtering. Magn Reson Med 1997, 37: 914–919. 10.1002/mrm.1910370617View ArticleGoogle Scholar
- Zijdenbos A, Dawant B, Margolin R, Palmer A: Morphometric analysis of white matter lesions in mr images: Method and validation. IEEE Trans Med Imag 1994,13(4):716–724. 10.1109/42.363096View ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.