Hill A., Thornham A., Taylor C., Model-Based Interpretation of 3D Medical Images. Fourth BMVC Conference, 1993, Guilford, England, pp.339-348.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....a set of orthogonal modes of variation . The training shapes are then represented by a subset of vectors which account for the majority of the observed variations. The PDM has proved extremely useful for image sequence analysis for tracking contours [53] and locating structures in medical images [58]. However, there was originally no intrinsic time dimension and, in early work, characteristic points for the training sets were selected by hand. More recent work by Baumberg and Hogg [54] successfully tackled both these drawbacks and has led to the development of not only tracking of walking ....

A. Hill, A. Thornham, and C.J. Taylor, \Model-based interpretation of 3D medical images," in British Machine Vision Conference, Guildford, UK, 1993, pp. 339-349.

....long term goal is to provide full 3D solutions. However, the increase in dimensionality multiplies the algorithmic and computational complexity. An intermediate solution could utilise 2D data in a 3D context. For example, a 3d model of the anatomy could be constructed from validated 2D structures [Hill et al. 1994a] Instead of registering images slice by slice, data interpolation is used to fill in the missing information [Barrett and Bess, 1994] of the 3D representation. Registration could then be carried out in 3D without the need for the selection of the correct target slice in the other image volume. ....

Hill, A., Thornham, A., and Talyor, C. J. (1994a). Model-based interpretation of 3D medical images. In Hancock, E. R., editor, Proceedings of the 5th British Machine Vision Conference, pages 339--48, York, UK. BMVA Press.

....the parameter vector representing the th i ST shape. Parameterisation is done using landmarks (other shape parameterisation methods may be utilized, e.g. Fourier descriptors [3] or B Splines [19] Landmarks are labelled either manually, as when a cardiologist labels the heart chamber boundaries [6,11], or (semi )automatically [10] Each landmark point is represented by its ) y x coordinate. Using L landmarks per frame and F frames per sequence, we can write the training set of ST shapes as 12 , N S SS S # K , where 12 ( ii iF i S iSppp ## K is the th i ST shape ....

Hill A., Thornham A., Taylor C., Model-Based Interpretation of 3D Medical Images. Fourth BMVC Conference, 1993, Guilford, England, pp.339-348.

....as more examples become available. This may be achieved by setting m 1 a . 4.3.2.7. Extension to 3D data The mathematical foundation of PDMs can be directly extended to 3D. What remains is the specific implementation details of the representation of 3D surfaces and the search algorithm used. In [23] the surfaces are represented by a set of 3D landmarks. A vector of length 3L is used to represent a surface with L landmarks and can be written as [ T L L L z y x z y x z y x , 2 2 2 1 1 1 u = x In the training stage of the 2D PDM as well as in 3D, we need to provide the coordinates ....

Hill A, Thornham A, Taylor C. Model-Based Interpretation of 3D Medical Images. 4th British machine Vision Conference, Guilford, England, 339-348, Sept. 1993.

....Then the image search is a problem of finding the set of parameters, for which the model template best fits the image data. If the amount of evidence found in the image data for a given instantiation of the model is described by an objective function (for examples of such function see [6] or [7]) the problem is one of optimisation of a multi modal, non linear function of many variables. Genetic algorithms are proposed as a method which is able to find good solutions (although not guaranteed to be optimal) to problems of this nature using very few trials [6] The main features of the ....

....movements derived for each point. The procedure is repeated until no significant change results. 3.3 Extension to 3D In the previous sections we presented the main features of Point Distribution Models and associated image search algorithms in 2D only. The 3D extension was introduced in [7]. The approach in 3D is essentially the same as in 2D. Each object is described by a labelled set of points (x 0 ; y 0 ; z 0 ; x 1 ; y 1 ; z 1 ; the model is trained on a set of examples; Genetic Algorithms and Active Shape Models are employed during the image search. Obviously, the ....

[Article contains additional citation context not shown here]

A. Hill, A. Thornham, and C. J. Taylor. Model-based interpretation of 3D medical images. In Proc. British Machine Vision Conference. BMVA Press, 1993.

....by averaging 305 brain in a stereotactic space. The variability is illustrated by the sharpness of the grey level contours, but the method does not provide suitable information to derive a shape model. Both methods are local and disregards the global nature of the shape variations. Hill et al. [6] performs a multivariate analysis on shape measurements derived from manually collected homologous points. Only objects having simple shapes and limited variations can be accounted. Therefore, the method is not applicable to modeling the whole brain. Martin et al. [8] presents a variant where ....

A. Hill, A. Thornham, and C.J. Taylor. Model-based interpretation of 3d medical images. In 4 th British Machine Vision Conference, pages 339--348, 1993.

....indicated a need for size and volume estimates of image objects. Examples where volume or size determination is useful include prostrate volume [43] ventricle volume [33, 39, 47] kidney volume [22] blood vessels [28] and fetus and placenta volumes [11] Segmentation and modelling of the data [2, 5, 8, 9, 10, 12, 13, 25, 26, 53, 54, 55, 56, 57] are also important research areas. Texture classification measurement of the statistical properties of material to aid diagnosis is another active research area [11, 61] 4 Conclusions 3 D free hand ultrasound imaging, compared to other 3 D ultrasound imaging techniques, has a number of ....

A. Hill, A. Thornham, and C. J. Taylor. Model-based interpretation of 3D medical images. In Proceedings of the British Machine Vision Conference, pages 339--348, Guildford, 1993.

....[4] utilise a set of static training shapes to derive a set of orthogonal modes of variation . The training shapes can be accurately represented by a basis consisting of a subset of these vectors. The PDM has proven useful in model based image interpretation (e.g. Cootes et al. [5] Hill et al. [6]) and in image sequence analysis (e.g. real time contour tracking [7] robust tracking of deformable models [8] However, one drawback of this approach is that there is no temporal aspect to the model. Hence it is not possible to extrapolate forward in time to get good estimates of the expected ....

Hill A., Thornham A., and Taylor C.J. Model-based interpretation of 3d medical images. In British Machine Vision Conference, volume 2, pages 339--349, 1993.

....camera) The boundary points are now reordered so that the #rst point is the reference point and approximated by a cubic B spline (for an example see #gure 3. 4) Each shape can be re#ected about its principal axis to double the volume of training data (as has been done by Hill, Thornham and Taylor [48]) 3.5.2 Approximating with a cubic B spline The control points of a length wise uniformly spaced B spline are used as a shape vector. Previous steps extract from each moving shape an ordered set of n boundary points W i = X i ; Y i ) with 0 i n which are approximated with a (closed) spline ....

A Hill, A Thornham, and C J Taylor. Model-based interpretation of 3d medical images. In British Machine Vision Conference, volume 2, pages 339#349, 1993.

....classified using a priori knowledge of object shape. Many objects are non rigid, and thus require some sort of deformable model in order to capture shape variability. One such model is the Point Distribution Model (PDM) 1] which has already been used as the basis for several vision applications [2, 3, 4, 5, 6]. An object is defined in terms of landmark points positioned strategically on various object features, and at regular intervals in between. By labelling such points on a set of training examples of the object, a statistical approach can be used to discover the mean object shape, and the major ....

A. Hill, A. Thornham, and C.J. Taylor. Model-based interpretation of 3D medical images. In Proc. BMVC, pages 339--348, Leeds, UK, 1992. Springer-Verlag.

....indicated a need for size and volume estimates of imaged objects. Examples where volume or size determination is useful include prostrate volume [36] ventricle volume [30, 33, 39] kidney volume [22] blood vessels [27] and fetus and placenta volumes [12] Segmentation and modelling of the data [1, 2, 5, 6, 7, 9, 13, 14, 24, 25, 43] are also important research areas. Texture classification measurement of the statistical properties of material to aid diagnosis is another active research area [12, 48] 2.6 Summary The investigation into the sources of error suggests how to best design a 3 D free hand ultrasound ....

A. Hill, A. Thornham, and C. J. Taylor. Model-based interpretation of 3D medical images. In Proceedings of the British Machine Vision Conference, pages 339--348, Guildford, 1993.

....the 3D hand PDM. 3 Tracking There has been much work on using PDMs for object location and tracking in both two and three dimensions. In most of this previous work, the dimensionality of the model has matched that of the input image (i.e. 2D model for 2D images [7, 8, 9] 3D model for 3D images [10]) Work on matching a 3D model to a 2D image has so far assumed a ground plane constraint and only one degree of rotational freedom [11, 12] We are attempting to match a 3D PDM to a 2D image under full 6 DOF. The key to model based object location is finding the set of model parameter values ....

A. Hill, A. Thornham, and C.J. Taylor. Modelbased interpretation of 3D medical images. In Proc. BMVC, pages 339--348, Guildford, UK, 1993. BMVA Press.

....In this paper we compare methods introduced by Brechbuhler[2] Kotche#[16] and [8] A fourth method is based on manually initialized subdivision surfaces similar to Wang[24] These methods are presented in more detail in sections 2.1 2.4. Similar approaches have also been proposed e.g. Hill[11] and Meier [18] Christensen[4] Szeliski[22] and Rueckert [20] describe conceptionally di#erent methods for warping the space in which the shapes are embedded. Models can then be built from the resulting deformation field [13, 9, 20] Brett[3] Rangarajan[19] and Tagare[23] proposed shape ....

Hill, A., Thornham ,A., Taylor ,C.J.: Model-Based Interpretation of 3D Medical Images. Brit. Mach. Vision Conf. BMCV, (1993) 339-348

....variation. They show how these Point Distribution Models (PDMs) can be used in image search [1,2] by creating Active Shape Models (ASMs) An ASM is analogous to a snake in that it refines its position and shape under the influence of image evidence, giving robust object location. Hill et al. [11] show how the PDM ASM approach can be extended to 3D when volume or range images are available, for example in medical imaging. A review of other deformable models is given in [1] Shen and Hogg [3] have recently shown how a fairly coarse flexible 3D model can be generated from a set of image ....

....cameras have seriously non square pixels, the factorisation method itself can give a good approximation of the true structure up to scaling [8,9] Thus we can use one of the reconstructed examples as a reference set. 6 Building a Flexible Model We build a 3D model using the method of Hill et al. [11]. We represent a set of 3D points x = x,y, z, 1) r (i = 1. n) as a single 3n element vector X = x,y, y,z, z) r Thus the set of N reconstructed objects in the reference frame are given by the N 3n element vectors Xj (1 = 1. N) We can calculate the mean of these, and ....

A.Hill, A.Thornham, C.J.Taylor. Model Based Interpretation of 3D Medical Images. in Proc. British Machine Vision Conference 1993. Vol.2. (Ed. J.Illingworth) BMVA Press. pp. 339-348.

....are iteratively updated to move the landmarks toward these better matched points, with the constraint that the overall shape cannot deform more than the examples seen in the training set. ASMs have been used successfully in a wide range of applications, including locating organs in medical images [11,12], face recognition and hand written character recognition [13] Our recent work has aimed at improving both the speed and the accuracy of the ASM method. In this paper we describe a multi resolution approach to modelling the grey levels around each landmark, and a coarseto fine strategy for ....

....of P are orthogonal and span the space of shape variations observed in the training set. If the shape parameters b are chosen inside suitable limits (derived from the training set) then the shapes generated by (1) will be similar to those in the training set. Examples of such models are given in [11 12]. In addition the local grey level environment about each landmark point can be modelled. Statistical information is gathered about the mean and covariances of the values of the pixels in the vicinity of each landmark, typically on profiles normal to the object boundary at that point. This data ....

A. Hill, A. Thornham and C.J.Taylor, Model-Based Interpretation of 3D Medical Images. in Proc. British Machine Vision Conference. Ed. J. Illingworth, BMVA Press, 1993, pp. 339-348.

....landmark, and find the main ways in which the shape tends to vary by examining the statistics of the point positions. This approach to modelling 3 D shape is an extension of earlier work on 2 D shape (Cootes et al. 12] which has also been used successfully with 3 D medical images (Hill et al. [13]) In our current work, however, the third dimension is the image intensity, so variations in this dimension cause the grey levels of the structures to change. We describe the model building approach, give some examples of models and show how they can be used to locate examples in new images. 2 ....

....data to calculate the . Each point becomes a triplet, xi, Yi, Ii) which can be considered as a point in a 3 D space. Each set ofn points describes the shape of a surface in this 3 D space. We have a set of examples, and can use the 3 D shape model similar to the one described by Hill et al. [13] to represent their mean and allowable vari ation. 2.2 Building 3 D Surface Models from Sets of Examples. The example shapes are aligned so that they overlap as much as possible. We allow a rigid rotation, scaling and translation in the x y plane, and a scaling and translation in the I ....

[Article contains additional citation context not shown here]

A. Hill, A. Thornham and C.J.Taylor, Model-Based Interpretation of 3D Medical Images. inProc. British Machine Vision Conference. Ed. J. Illingworth, BMVA Press, 1993, pp. 339-348.

....of an existing framework for establishing dense correspondences between a set of training examples [4] to build a 3D Point Distribution Model. Examples are given for both synthetic and real data. 1 Introduction We are interested in building Point Distribution Models (PDMs) of 3D shapes [6]. This requires dense correspondences to be established between a set of training examples of the shape. Previous publications [3] including some describing methods of surface correspondence [7] 1] have described probable solutions to parts of this problem. Here we describe a completely automated ....

A. Hill, A. Thornham, and C. J. Taylor. Model-based interpretation of 3D medical images. In J. Illingworth, editor, # ## British Machine Vison Conference, pages 339--348, Guildford, England, Sept. 1993. BMVA Press.

.... Model (PDM) 3] Such a model may be combined with statistical models of the greyscale appearance of the modelled shape within training images to produce an Active Shape Model (ASM) 4] ASMs have been shown to perform accurately and robustly in the segmentation of medical images in 2D [13] and 3D [8]. Currently, the construction of PDMs involves the manual identification of a set of L landmarks fx i ; 1 i Lg for each of N training examples of shape. A landmark is a point which identifies a salient feature of the shape and which is present on every example of the class. Manual definition of ....

A. Hill, A. Thornham, and C. J. Taylor. Model-based interpretation of 3d medical images. In J. Illingworth, editor, 4 th British Machine Vison Conference, pages 339--348, Guildford, England, Sept. 1993. BMVA Press.

....in which the point should move to best fit the image evidence. Evidence from all the landmark points is combined to calculate an overall deformation of the shape in order to produce a better model fit. To date, most of the work on ASMs has been in 2D. The first implementation of ASMs in 3D [6] has shown that they work well for the automated segmentation of 3D grey scale images. However, the problems of image navigation and region definition in 3D make model building difficult and this implementation was unsuitable for routine use. In addition, surfaces could only be built from closed ....

Andrew Hill, Ann Thornham, and Christopher J. Taylor. Model-Based Interpretation of 3D Medical Images. In John Illingworth, editor, 4 th British Machine Vision Conference, pages 339--348, Guildford, England, 1993. BMVA Press.

....to add extra examples to the training set. However, making 2D measurements on slices taken from 3D images is prone to errors due to differences in slice orientation between images. To eliminate these errors, measurements need to be made in 3D. We are currently investigating the use of a 3D ASM [25] to make a full 3D assessment of cartilage thickness. This is likely to be a more informative measurement than cartilage volume [13] which will allow us to automatically monitor early focal changes in cartilage thickness, enabling large scale use of the technique in clinical trials. ....

A. Hill, A. Thornham, C. J. Taylor, Model-Based Interpretation of 3D Medical Images. In 4 th British Machine Vision Conference (Ed. J. Illingworth), 339--348, Guildford, England (1993), BMVA Press.

....extension of an existing framework for establishing dense correspondences between a set of training examples [4] to build a 3D Point Distribution Model. Examples are given for both synthetic and real data. 1 Introduction We are interested in building Point Distribution Models (PDMs) of 3D shapes [6]. This requires dense correspondences to be established between a set of training examples of the shape. Previous publications [3] including some describing methods of surface correspondence [7] 1] have described probable solutions to parts of this problem. Here we describe a completely automated ....

A. Hill, A. Thornham, and C. J. Taylor. Model-based interpretation of 3D medical images. In J. Illingworth, editor, 4 th British Machine Vison Conference, pages 339--348, Guildford, England, Sept. 1993. BMVA Press.

No context found.

Hill A., Thornham A., Taylor C., Model-Based Interpretation of 3D Medical Images. Fourth BMVC Conference, 1993, Guilford, England, pp.339-348.

No context found.

A. Hill, A. Thornham, and C. Taylor, \Model-based interpretation of 3d medical images," in British Machine Vision Conference, Springer-Verlag, ed., pp. 339-348, 1992.

No context found.

Hill A, Thornman A, Taylor CJ. Model-based interpretation of 3D medical images. Proceedings of the British Machine Vision Conference, England: Guildford; 1993. 339-348.

No context found.

A. Hill, A. Thornham, and C. J. Taylor. Model-based interpretation of 3D medical images. In British Machine Vision Conference, pages 339--348, 1993.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC