## Segmentation of the liver using a 3D statistical shape model (2004)

Citations: | 16 - 2 self |

### BibTeX

@TECHREPORT{Lamecker04segmentationof,

author = {Hans Lamecker and Thomas Lange and Martin Seebaß},

title = { Segmentation of the liver using a 3D statistical shape model},

institution = {},

year = {2004}

}

### OpenURL

### Abstract

This paper presents an automatic approach for segmentation of the liver from computer tomography (CT) images based on a 3D statistical shape model. Segmentation of the liver is an important prerequisite in liver surgery planning. One of the major challenges in building a 3D shape model from a training set of segmented instances of an object is the determination of the correspondence between different surfaces. We propose to use a geometric approach that is based on minimizing the distortion of the correspondence mapping between two different surfaces. For the adaption of the shape model to the image data a profile model based on the grey value appearance of the liver and its surrounding tissues in contrast enhanced CT data was developed. The robustness of this method results from a previous nonlinear diffusion filtering of the image data. Special focus is turned to the quantitative evaluation of the segmentation process. Several

### Citations

2232 |
A method for registration of 3D shapes
- Besl, McKay
- 1992
(Show Context)
Citation Context ... M E j= x ∈S M' d(xj,S') 2 + y d(xj,S) 2 We choose an iterative optimization strategy, where at each iteration we first compute optimal rigid transformation parameters T by applying the ICP-algorithm =-=[33]-=-. We then perform the optimization with respect to the shape parameters b. This procedure is repeated until convergence. By separating the optimization (12) for the shape and transformation parameters... |

1506 | C.: Active appearance models
- Cootes, Edwards, et al.
- 1998
(Show Context)
Citation Context ...te shape (liver from visible-human data set) in the deformation process. Only few examples have been quantitatively evaluated in this work. Beichel et al. [8] applied an Active Appearance Model (AAM) =-=[9]-=- to the segmentation of the diaphragm dome which separates the lungs and the heart from the liver. Only the upper part and not the whole liver is segmented. They model the 3D shape of the dome by mean... |

514 | W.: multi-resolution analysis of arbitrary meshes
- Eck, DeRose, et al.
- 1999
(Show Context)
Citation Context ... constraint, we require the boundary of S1 to be mapped smoothly to the boundary of D. One possible candidate for a map f that minimizes metric distortion could be piecewise linear (PL) harmonic maps =-=[29]-=-. However these maps may introduce fold-overs, as was shown in [19]. An alternative, which is guaranteed to be fold-over free, are so-called convex-combination maps [21]. Without loss of generality, l... |

455 | Deformable models in medical image analysis: A survey. Medical image analysis
- McInerney, Terzopoulos
- 1996
(Show Context)
Citation Context ...er reduction is still desirable. Among the vast variety of automatic image segmentation approaches deformable models have proven to be effective and powerful in a large number of medical applications =-=[6]-=-. In many cases simple assumptions on the deformability of the object, like smoothness or elasticity criteria, generate shapes that definitely are not legal instances of the object. Including a-priori... |

292 | C.J.Taylor and J.Haslam, "The Use of Active Shape Models For Locating Structures in Medical Images
- Cootes
- 1993
(Show Context)
Citation Context ...stance between liver and skin). In our work we will follow the approach of capturing the shape variations of the liver from a set of training data statistically via principal component analysis (PCA) =-=[11]-=-. The deformable model is only allowed to vary within the captured space of variations. This approach seems to be promising for robust and automatic image segmentation, because it provides an efficien... |

254 | Parametrization and smooth approximation of surface triangulations
- FLOATER
- 1997
(Show Context)
Citation Context ...re as follows. First, we adopt the optimality criterion of minimal geometric distortion for shape correspondence, as used by Brett et al. [18]. Instead of harmonic maps we use convex-combination maps =-=[21]-=- which are guaranteed to be fold-over free. This idea is combined with a morphing method by Z¨ockler et al. [22] which is capable of matching shapes of arbitrary topology. This approach also allows ex... |

181 |
Elementary Differential Geometry
- O’Neill
- 1966
(Show Context)
Citation Context ...eighbors. Yet we are interested in producing a mapping, that preserves the metric structure of the surface S1 as good as possible. We use here the idea of locally approximating the geodesic polar map =-=[30]-=-, first presented by Floater [21]. Thereby λij are computed as convex weights in the planar domain, such that the angles and lengths of the original triangulation are preserved as good as possible (se... |

179 | A minimum description length approach to statistical shape modelling
- Davies, Twining, et al.
(Show Context)
Citation Context ...n planar domains using harmonic maps. Their method applies only to surfaces homeomorphic to discs and does not guarantee foldover free mappings however (see [19] for an illustration of this fact). In =-=[20]-=- the correspondences are defined using an information theoretic minimal description length approach. Here the optimality criterion is the compactness of the model. The optimization of this criterion i... |

170 | Efficient and reliable schemes for nonlinear diffusion filtering
- Weickert, Romeny, et al.
- 1998
(Show Context)
Citation Context ...is a smoothed version of u which is obtained by convolving u with a Gaussian of standard deviation σ. Equation 5 can efficiently be solved by using the additive operator splitting scheme suggested in =-=[32]-=-. For this work we implemented 2D diffusion filtering and applied the filter slice by slice to the image data, using the values σ = 3.0, λ = 3.5 and t = 100. The effect of the diffusion filter on a CT... |

159 | Shape transformation using variational implicit functions
- Turk, O’Brien
- 1999
(Show Context)
Citation Context ... intermediate slices between each two consecutive slices of the original image stack, such that the resulting voxel size is nearly isotropic. The method of variational implicit function interpolation =-=[24]-=- is well suited for this task. For reasons of efficiency all surfaces are simplified [25] by reducing the number of triangles, obtaining meshes with about 25000 triangles and 12500 nodes. All these st... |

101 |
Statistical Factor Analysis and Related Methods. Theory and Applications
- Basilevsky
- 1994
(Show Context)
Citation Context ...del A statistical model is built from a training set of shapes vi (i = 1,..., N). Each shape vi is given by M points sampled on its surface (thus vi ∈ R ). Using standard principal component analysis =-=[27]-=- each shape vector can be expressed using a linear model of the form vi = v + Pbi = v+\pb i where v is the mean shape vector and P = {p } the matrix of eigenvectors of the covariance matrix. The corre... |

81 | A Review of Nonlinear Diffusion Filtering
- Weickert
- 1997
(Show Context)
Citation Context ... In a preprocessing step the CT data are smoothed with a nonlinear isotropic diffusion filter to reduce noise but preserve the liver boundary. An overview of nonlinear diffusion filtering is given in =-=[31]-=-. Let f(x) be the image then a filtered image u(x, t) is calculated by solving the partial differential equation The diffusivity function tu = div (g(|^7ua| )Vu) with u(x,0) = f(x). (5) I 1 (s < 0) g(... |

60 | Shape-based 3D surface correspondence using geodesics and local geometry
- Wang, Peterson, et al.
- 2000
(Show Context)
Citation Context ...ly difficult in 3D. Several approaches have been proposed to face the problem in 3D and to automate the task: 3sA method for 3D shape correspondence using local geometry and geodesics is described in =-=[12]-=- without its application to statistical shape models. It is evaluated by computing residual surface distances. In [13] the shapes are represented by their expansion into a series of elliptical harmoni... |

57 | Automatic construction of multiple-object three-dimensional statistical shape models: application to cardiac modeling
- Frangi, Rueckert, et al.
(Show Context)
Citation Context ... Fleute et al. [15] establish correspondence by elastic registration of a template shape with all other shapes based on minimizing Euclidean distance. Closely related is the approach of Frangi et al. =-=[16]-=-. The PCA is performed on the points of the control grids computed from an elastic registration on binary volumetric data. Brett et al. [17, 18] present two approaches of automatic construction of sha... |

45 |
Toga, “Detection, visualization and animation of abnormal anatomic structure with a deformable brain atlas based on random vector field transformation,” Med
- Thompson, W
- 1997
(Show Context)
Citation Context ...lliptical harmonics. The PCA is performed on the parametric description in contrast to the point distribution model used in [11]. Only shapes with sphere-like topology can be treated. Thompson et al. =-=[14]-=- compute a mapping from each shape onto a sphere using a deformable model approach. Correspondence between two spheres, constrained by matched anatomical feature lines, is established by means of a wa... |

34 | Volumetric medical images segmentation using shape constrained deformable models., CVRMed-MRCAS
- Montagnat, Delingette
- 1996
(Show Context)
Citation Context ...be sensible to increase robustness. There are only few approaches so far that use 3D deformable models for the segmentation of the liver. In [3] a globally constrained 3D deformable model proposed in =-=[7]-=- is applied to the segmentation of the liver. This method incorporates prior knowledge about the curvature of one template shape (liver from visible-human data set) in the deformation process. Only fe... |

33 | Hierarchical computation of PL harmonic embeddings
- Duchamp, Certain, et al.
- 1997
(Show Context)
Citation Context ...roach of parameterizing surfaces onto common planar domains using harmonic maps. Their method applies only to surfaces homeomorphic to discs and does not guarantee foldover free mappings however (see =-=[19]-=- for an illustration of this fact). In [20] the correspondences are defined using an information theoretic minimal description length approach. Here the optimality criterion is the compactness of the ... |

31 | Efficient semi-automatic segmentation of 3D objects in medical images
- Schenk, Prause, et al.
(Show Context)
Citation Context ...asing number of slices from improved CT scanners. Therefore methods have to be devised that require few interaction time. A semi-automatic method, that reduces manual interaction time is presented in =-=[5]-=-. In this approach only a few slices of the original CT stack are interactively segmented using a live-wire approach, and for the intermediate slices a shape based interpolation is performed automatic... |

31 | A method of automated landmark generation for automated 3D PDM construction
- Brett, Taylor
- 1998
(Show Context)
Citation Context ... distance. Closely related is the approach of Frangi et al. [16]. The PCA is performed on the points of the control grids computed from an elastic registration on binary volumetric data. Brett et al. =-=[17, 18]-=- present two approaches of automatic construction of shape models based on polyhedral shape representation. One minimizes the Euclidean distance between shapes on different levels of detail, obtained ... |

29 |
Fully automatic anatomical, pathological, and functional segmentation from CT scans for hepatic surgery
- Soler, Delingette
(Show Context)
Citation Context ...f the liver tissue. The segmentation is the basis for a 3D anatomical model of the liver, which permits the computation of the resection volume, i.e. the tissue to be removed during surgery [1], [2], =-=[3]-=-. This volume contains the tumor and the part of the liver parenchyma where no blood supply would be provided after the resection. The high variability of the liver shape and vasculature [4] requires ... |

25 | H.C.: Fast and intuitive generation of geometric shape transitions
- Zockler, Stalling, et al.
- 2000
(Show Context)
Citation Context ..., as used by Brett et al. [18]. Instead of harmonic maps we use convex-combination maps [21] which are guaranteed to be fold-over free. This idea is combined with a morphing method by Z¨ockler et al. =-=[22]-=- which is capable of matching shapes of arbitrary topology. This approach also allows experts to manually specify corresponding features on two shapes, thereby constraining the correspondence mapping.... |

20 |
Surface simplification using quadratic error metrics
- Garland, Heckbert
- 1997
(Show Context)
Citation Context ...h that the resulting voxel size is nearly isotropic. The method of variational implicit function interpolation [24] is well suited for this task. For reasons of efficiency all surfaces are simplified =-=[25]-=- by reducing the number of triangles, obtaining meshes with about 25000 triangles and 12500 nodes. All these steps are performed with the help of the visualization and modeling software Amira [26]. 2.... |

16 | Three-dimensional Model-based Segmentation of Brain MRI
- Kelemen, Székely, et al.
- 1998
(Show Context)
Citation Context ...od for 3D shape correspondence using local geometry and geodesics is described in [12] without its application to statistical shape models. It is evaluated by computing residual surface distances. In =-=[13]-=- the shapes are represented by their expansion into a series of elliptical harmonics. The PCA is performed on the parametric description in contrast to the point distribution model used in [11]. Only ... |

16 | M.: A Statistical Shape Model for the Liver
- Lamecker, Lange, et al.
- 2002
(Show Context)
Citation Context ...his method to the generation of statistical shape models. Second, a statistical shape model of the liver is built from a training set of 43 shapes, extending preliminary work presented by the authors =-=[23]-=-. This is first 3D statistical shape model of the liver applied to the task of image segmentation. Third, the model is applied to the automatic segmentation of 33 CT data sets and the results are eval... |

13 |
Automatic 3d segmentation of the liver from abdominal ct images: a level-set approach
- Pan, Dawant
- 2001
(Show Context)
Citation Context ... 3D shape of the dome by means of a 2D AAM, where one dimension (z-direction) is represented by a height grey value image above a reference plane. A 2D and 3D level-set approach is used by Pan et al. =-=[10]-=- to segment the liver. They proposed a novel speed function that controls the front propagation of an implicitly defined surface towards the liver boundary. The idea is to change the speed function dy... |

13 |
Incorporating a statistically based shape model into a system for computer assisted anterior cruciate ligament surgery”, Med Imaging Anal
- Fleute, Lavalleé, et al.
- 1999
(Show Context)
Citation Context ...ape onto a sphere using a deformable model approach. Correspondence between two spheres, constrained by matched anatomical feature lines, is established by means of a warping algorithm. Fleute et al. =-=[15]-=- establish correspondence by elastic registration of a template shape with all other shapes based on minimizing Euclidean distance. Closely related is the approach of Frangi et al. [16]. The PCA is pe... |

7 |
H.C.: Amira - an object oriented system for visual data analysis
- Stalling, Westerhoff, et al.
- 2005
(Show Context)
Citation Context ...ied [25] by reducing the number of triangles, obtaining meshes with about 25000 triangles and 12500 nodes. All these steps are performed with the help of the visualization and modeling software Amira =-=[26]-=-. 2.2 Building the Statistical Model A statistical model is built from a training set of shapes vi (i = 1,..., N). Each shape vi is given by M points sampled on its surface (thus vi ∈ R ). Using stand... |

6 |
Virtual planning of liver resections: image processing, visualisation and volumetric evaluation
- Glombitza, Lamad, et al.
- 1999
(Show Context)
Citation Context ...entation of the liver tissue. The segmentation is the basis for a 3D anatomical model of the liver, which permits the computation of the resection volume, i.e. the tissue to be removed during surgery =-=[1]-=-, [2], [3]. This volume contains the tumor and the part of the liver parenchyma where no blood supply would be provided after the resection. The high variability of the liver shape and vasculature [4]... |

4 |
et al.: Segmental anatomy of the liver: poor correlation with CT
- Fasel, Selle, et al.
- 1998
(Show Context)
Citation Context ...[1], [2], [3]. This volume contains the tumor and the part of the liver parenchyma where no blood supply would be provided after the resection. The high variability of the liver shape and vasculature =-=[4]-=- requires individual planning. In the clinical routine the manual segmentation of the liver parenchyma is too time consuming (about one hour per data set). The segmentation time will be even longer wi... |

3 | and appearance-based segmentation of volumetric medical images
- Beichel, Mitchell, et al.
- 2001
(Show Context)
Citation Context ...r knowledge about the curvature of one template shape (liver from visible-human data set) in the deformation process. Only few examples have been quantitatively evaluated in this work. Beichel et al. =-=[8]-=- applied an Active Appearance Model (AAM) [9] to the segmentation of the diaphragm dome which separates the lungs and the heart from the liver. Only the upper part and not the whole liver is segmented... |

3 | Least squares fitting of two structures - McLachlan - 1979 |

2 |
construction of 3D shape models using harmonic maps,” in Medical Image Understanding and Analysis
- “Automated
- 2000
(Show Context)
Citation Context ... distance. Closely related is the approach of Frangi et al. [16]. The PCA is performed on the points of the control grids computed from an elastic registration on binary volumetric data. Brett et al. =-=[17, 18]-=- present two approaches of automatic construction of shape models based on polyhedral shape representation. One minimizes the Euclidean distance between shapes on different levels of detail, obtained ... |