Anatomical models in cardiovascular image analysis

(1)

(2)

(3)

(4)

Anatomical models in cardiovascular

image analysis

PROEFSCHRIFT

ter verkrijging van de graad van Doctor aan de Universiteit Leiden,

op gezag van de Rector Magnificus Dr. W.A. Wagenaar, hoogleraar in de faculteit der Sociale Wetenschappen,

volgens besluit van het College voor Promoties te verdedigen op donderdag 21 oktober 1999

te klokke 15.15 uur

door

Boudewijn Pieter Frederik Lelieveldt geboren te Leiden

(5)

Promotiecommissie:

Promotor: Prof. dr. ir. J.H.C. Reiber

Referent: Prof. M. Sonka, Ph.D (The University of Iowa, Iowa City, Iowa, USA)

Overige leden: Prof. dr. E.E. van der Wall Prof. dr. A. de Roos Dr. ir. J. Dijkstra

(6)

CIP-GEGEVENS KONINKLIJKE BIBLIOTHEEK, DEN HAAG

© Boudewijn Lelieveldt. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanic, pho-tocopying, recording or otherwise, without prior permission of the author.

ISBN 90-9012908-1 NUGI 743

Lay-out by J.J. Beentjes.

(7)

(8)

1

Introduction

Medical imaging techniques play an essential role in medicine for diagnostic, treatment planning and patient monitoring purposes. Over the last century, imaging modalities such as X-ray imaging and much later ultrasound imaging have become available to the clinician to examine a patient's condition and the changes therein over time (either with or without an intervention). The last two decades have shown a significant increase in the development and use of three-dimensional imaging techniques like Computed X-ray Tomography (CT) and Magnetic Resonance Imaging (MRI). Particularly MRI has rap-idly become very popular for its flexibility and clinical versatility, as anatomical informa-tion can be acquired in combinainforma-tion with funcinforma-tional data. A number of specialized MR acquisition methods is available to depict vessel structures like peripheral and coronary vessels (Magnetic Resonance Angiography, MRA), measure blood flow velocities (Mag-netic Resonance Velocity Mapping), examine metabolic processes (MR Spectroscopy) or measure myocardial wall contraction properties (MR tagging, MR velocity imaging, multi-slice multi-phase imaging). Especially in cardiology, MRI may potentially replace different types of other imaging modalities; for that reason MRI is increasingly referred to as the 'one-stop shop' approach for cardiac imaging.

The human observer plays a pivotal role in the processing and analysis of medical image data. In performing an image interpretation task, the human eye draws from a battery of intellectual resources in the form of prior knowledge, experience or expecta-tion. As a result of this, a human observer is superbly capable of localizing organs in a medical image, analyzing organ dimensions and distinguishing normal from abnormal anatomical features. Unfortunately, the human eye is not fit to perform accurate mea-surements in medical images, and in clinical practice many diagnostic and treatment decisions are taken on the basis of visual estimates of the severity of a condition.

(11)

computer is supplementary to the human eye in the sense that it can provide accurate quantitative measures based on local image information.

To derive measurements from image data, it is often necessary to outline the struc-ture(s) of interest in the data, a process referred to as image segmentation. In many clini-cal applications, this is performed manually or semi-automaticlini-cally, simply because the computer is not capable of interpreting an image scene autonomously. Therefore, com-puter-aided measurements are generally performed under human supervision, where an observer performs the complex task of image interpretation and the computer provides quantitative measurements. Furthermore, the human operator evaluates the computer output, and intervenes in case the automated segmentation results contradict human perception and common sense.

With the advent of tomographic imaging techniques like MRI and CT, which gener-ally provide three- or higher dimensional image data, the amount of data per examina-tion has increased to such an extent that manual analysis has become inpractical. For example, a cardiac MR examination provides the radiologist or cardiologist with a fully three-dimensional visualization of the heart in various stages in the cardiac cycle (typi-cally 10 slices x 20 phases at each stage, totaling ± 200 images per examination). The time required to accurately analyze such image sets manually (2-3 hours) is a major obstacle for the widespread use of MRI as a routine diagnostic tool to quantitatively assess cardiac function. Secondly, manual contour drawing is a significant source of inter- and intra observer variabilities in clinical practice. These two factors have created a great demand for reliable tools to further automate the segmentation and functional quantification of medical image data.

A vast body of work has been described on image segmentation and its applications in medicine, where many automated segmentation methods merely utilize low-level image data. Unfortunately, the success rate of such data-driven segmentation methods greatly depends on the application domain. In MR images of the brain, the image conditions and tissue contrasts are such that a fully automatic segmentation is within reach purely based on low-level image features. However, most imaging protocols and modalities are susceptible to variations in image quality, where information may be unreliable or miss-ing. In many cases, the human eye is still capable of tracking a particular organ in the image data based on experience and prior knowledge, where a data driven segmentation method clearly fails. There is a growing consensus in the medical imaging community that the integration of prior knowledge into data driven approaches is of critical impor-tance to improve the robustness of automated segmentation methods.

(12)

Introduction 13 examining its shape. In each stage a tradeoff between generality and accuracy is made, where in the first stage emphasis lies on generality and in the second stage on accuracy. The scope of the work described in this thesis is the development of knowledge repre-sentations suitable to further automate the highest level in the image interpretation chain for cardiovascular MR-images. By providing the computer with a coarse model of what different organs look like in their spatial context in cardiac MR-images, a segmentation problem can be approached in a similar coarse-to-fine manner as occurs in the process of human perception. The anatomical models developed in this work provide the computer with knowledge about ‘the general picture’ of thoracic anatomy as it appears in two types of cardiovascular MR-images with an emphasis on generality. They are intended to auto-matically provide the initial conditions for locally accurate segmentation methods, and are therefore not intended to be accurate in itself.

(13)

(14)

2

Anatomical models in medical

image analysis

Anatomical modeling is a rapidly growing field of research, where three main application areas have evolved over the last decades: visualization and education, functional analysis and segmentation. Since the focus of attention in this thesis is directed to anatomical models for segmentation purposes, this section will briefly discuss the first two applica-tions, whereas models aimed at medical image segmentation are discussed in more detail.

2.1 Anatomical models for visualization and educational purposes

Recent advances in computer technology have enabled the development of digital tomical models for education and training purposes. For example, highly detailed ana-tomical atlas models are digitally represented in the form of a set of labeled voxels [1-9]. By combining photorealistic renderings of segmented voxel data with background infor-mation about the visualized structures, different aspects of human anatomy can be explored three-dimensionally. The added value of such atlases over plain paper atlases is the facility to interactively explore the spatial structure of an organ in three dimensions, where background information about the function of an organ can be retrieved on demand. A well-known example of such an atlas model for visualization and education purposes is the visible human atlas [4-9], where human anatomy can be viewed interac-tively in combination with its appearance in a number of radiological imaging modali-ties.

(15)

[13, 14]. For a recent overview of the state-of-the-art applications of anatomical models in surgical simulation applications the reader is referred to [15].

A third promising application of anatomical atlas models in combination with visual-ization techniques lies in radiological image simulation. In this application, the appear-ance of different organs in a radiological image is simulated by modeling the entire imaging chain from physical tissue characteristics to the underlying physical principles and transfer functions of an imaging modality. This way, the role of specific parameters in the imaging chain on the resulting image conditions can be investigated. Examples of this application of anatomical atlas models have been described in [16-19] for nuclear imaging and in [20, 21] for magnetic resonance images.

2.2 Anatomical models for functional analysis

A second important application field of anatomical models in medicine is the analysis and simulation of physical processes as they occur throughout the human body. By cou-pling physiological knowledge with image data, a better insight can be obtained in the features that distinguish normal from pathological conditions. An excellent example of this application is provided by Kaye [22], who modeled cardiothoracic interactions dur-ing respiration and the changes therein as a result of a pneumothorax based on an ana-tomical model of the thoracic contents. Other examples of the applications of anatomical models for functional analysis are given in [23, 24].

A specialized application area of analytical anatomical models is formed by the dynamic cardiac deformation models [25-63], which are aimed at modeling the contrac-tion pattern of left [27-62] and right [63] ventricles over the cardiac cycle. These defor-mation models are dedicated to the tracking and analysis of wall motion, often by means of a set of characteristic parameter functions, from which wall motion abnormalities can be recognized as deviations from a normal set of functions (e.g.[27, 42, 43]). Further-more, regional parameters for cardiac function like wall stress and strain can be estimated from the deformations mapping the model to an image set.

(16)

Anatomical models in medical image analysis 17 2.3 Anatomical models for medical image segmentation

The need to incorporate prior knowledge into image segmentation methods is nowadays widely recognized. Especially in medical imaging, where many aspects of the imaging conditions are difficult to control, the incorporation of knowledge about the shape, loca-tion, appearance and spatial context of an organ is essential. In this secloca-tion, different classes of anatomical models are discussed in the context of a commonly applied subdivi-sion of segmentation methods into five abstraction levels. From high-level down to low-level operations, one can distinguish the scene low-level, the single object low-level, the image entity level, the low-level segmentation level and the preprocessing level (see Figure 2.1).

Each stage in the image interpretation hierarchy is classified according to the degree of prior knowledge involved in operations at that level. The different anatomical modeling methods described in the literature can be placed mainly at the top three levels of the image interpretation pyramid, and are discussed in a bottom-up order.

Physically-based deformable models (snakes):

A widely acknowledged object representation applied for segmentation of medical imag-ery is the deformable model, which has been recently surveyed in Singh [77]. Terzopou-los [78] introduced the deformable model concept to create realistic computer

animations by viewing an object surface as an elastic sheet and deforming the object by mimicking the physical deformation behavior of the sheet. The first application to image

Preprocessing Low-level image analysis Lower image interpretation

Raw images Filtered images

Image features: edges,

regions, texture Scene elements: 3D-surfaces,

volumes, contours Objects

Object recognition High level scene interpretation Scene

(17)

segmentation was described by Kass [79], who introduced the nowadays widely applied active contours, also referred to as snakes. Roughly, three types of snakes have evolved since then: parametric snakes, implicit snakes and probabilistic snakes.

Parametric snakes

In brief, a parametric snake is a curve expressed in coordinate functions (x(s) and y(s)), where s represents the parametric domain [0,1]. The shape of the contour is governed by an energy functional:

(2.1)

where the first integral term represents an internal deformation energy of the model, which is balanced with an external scalar field P(v), typically defined from an image fea-ture such as the local image gradient. Parameter functions w1(s) and w2(s) represent two

physical properties of the contour, i.e. the ability to stretch and bend respectively. These functions can be used to impose a preferred shape on the model and to locally control shape characteristics like the object smoothness of the resulting segmentation. The third term Euser represents shape constraints introduced by the user, e.g. by fixating a point of the contour to an image point. The final shape of an active contour in an image corre-sponds to a minimum in E(v), which can be found by numerically solving the Euler-Lagrange equation of Equation 2.1.

Extensions of the two-dimensional snake model to three dimensions (deformable bal-loons) have since been reported [49, 80-94], as well as many modifications of the origi-nal energy formulation to improve robustness of the snake- and balloon methods with respect to spurious feature points and initial positioning [81, 95, 96], transitions in topology [97, 98] and simultaneous detection of multiple objects [99]. Applications to left-ventricular segmentation in dynamic cardiac MR-images are given in e.g. [53, 62, 84, 100-104]

Implicit snakes

One of the practical limitations of parametric snakes is the requirement of an initial guess, which is reasonably close to the desired shape. Furthermore, these snakes are not suitable to describe shape protrusions or extrusions that a shape may posses. A different class of physically-based deformable models designed to circumvent these shortcomings is the level-set approach [105-111], which simulates the propagation of wave fronts with curvature dependent speeds. In [108, 109], the original formulation of Cassalles [105, 106] is modified to an energy minimization, whereas in [111, 112] an extension for simultaneous segmentation of multiple objects is described. An advantage of these implicit snakes over the parametric snake formulation is the lack of assumptions made about topological structure of an object. Therefore topologically adaptable snakes have shown to be useful for segmentation of complexly shaped objects of which little prior shape knowledge is available, for instance branching vessel structures. However, in the

(18)

Anatomical models in medical image analysis 19 absence of geometric shape constraints other than connectivity, the ways to incorporate shape knowledge are limited in case prior knowledge is available.

Probabilistic snakes

Due to the locally distributed nature of prior shape knowledge of the original parametric and implicit snake formulations, the facilities to incorporate shape knowledge other than smoothness constraints to restrict the allowable shape domain are limited. Therefore these snakes are less suitable for object recognition purposes and can be generally classi-fied to the low-level image interpretation stage in Figure 2.1. As a result of this, paramet-ric and implicit snakes are mainly suitable for application in a highly interactive setting.

By selecting a shape parametrization expressed on an orthonormal basis, i.e. a repre-sentation that allows the definition of an object shape as a weighted sum of known basis functions, the shape parameters become physically interpretable. A preferred shape can thus be imposed, based on the parameter distributions over a set of training samples, where the snake is allowed to deform following population-based shape deformations. The probabilistic snake is preferentially attracted towards feature patterns in the image data, which are consistent with its trained shape. This makes the model matching more robust with respect to initial position and noise and therefore these snakes can be classi-fied to the object recognition level in Figure 2.1.

Applications of probabilistic snakes have been described in Vemuri et al. [113-115], who developed a model based on a deformable superquadric in combination with a locally superimposed deformation field expressed on an orthonormal wavelet basis. Staib et al. describe a two-dimensional probabilistic snake [116] and a three-dimensional [103] probabilistic balloon model applicable to deformations of four surface topologies constructed on a Fourier basis, whereas orthonormal Fourier parametrizations have been described in [117-119] that are applicable to segmentation and recognition problems of free-form closed surfaces.

Statistical shape models

Probabilistic snakes describe a shape and its natural variations by means of the parameter distributions of a shape parametrization and have shown to be a powerful representation for population based shape knowledge. The necessary prerequisite of a parametrization on an orthonormal basis however introduces limitations to the shape topology of these models. In contrast, statistical shape models do not require a shape parametrization, and are therefore not subjected to the topological or shape constraints intrinsic to a lumped parameter model. This makes these statistical models suitable to describe free-form shapes consisting of multiple objects simultaneously.

(19)

variations around a shape average can be extracted by means of a principal component analysis on the sample point distributions. The only necessary condition for the calcula-tion of a point distribucalcula-tion model is the definicalcula-tion of a point-correspondence between points in successive training samples, which ensures a compact and specific model. In two dimensions this point correspondence is typically defined in application specific assumptions, which are difficult to generalize in three dimensions. The development of more generic methods to define point correspondence for two [126-128] and three [129, 130] dimensions is currently an active field of research.

The application of point distribution models to image segmentation is known as Active Shape Model (ASM). A key difference between the ASM matching method and the matching mechanism for parametric and implicit snake models is the absence of an energy functional based on elastic material properties. For ASM’s, the image matching is performed by calculating a suggested boundary location for each point in the PDM based on image information. This allows an elegant coupling between high-level knowl-edge about object shapes and low-level image features. The suggested boundary hypoth-eses can be generated either using a simple edge filter, but also using custom edge filters for each point in the shape samples, a statistical gray-value model for each sample point or other forms of prior knowledge about an organ’s image appearance [122, 131-133]. The model pose- and shape parameters are iteratively updated to optimally fit the hypothesized shape, where the model is only allowed to deform along the most charac-teristic eigendeformations. A final solution is reached when the generated candidate boundary points coincide with the model boundaries.

A second important statistical shape modeling method for segmentation purposes is based on spatial normalization of a set of training images (2D) or image volumes (3D). By optimally registering a set of segmented voxel volumes into a standardized space by applying affine [134-140] or higher dimensional transformations such as thin-plate spline interpolants [141-143], an image scene can be expressed as a probability map or as an average shape with a locally defined variance measure respectively. These models are generally applied to segmentation problems by weighting a feature-based probability density function with a spatial probability distribution of an organ shape [138, 139, 141-143] in a Bayesian formulation.

Because statistical shape models allow a coupling between low-level image data and higher level knowledge about individual organ shapes and their spatial context in a scene, they can be classified to the scene interpretation level in the image interpretation hierarchy in Figure 2.1.

Boundary template models

(20)

primi-Anatomical models in medical image analysis 21 tives, as has been demonstrated in 2D by Yuille[152], who describes the eye and the mouth as a combination of circles and parabolic arcs. In 3D this analytical approach has been exemplified by Delibasis [153], who modeled part of the brain stem as a combina-tion of globally deformable superquadrics. Furthermore, in Chapters 4, 5 and 6 of this thesis [154-156], a novel boundary template model is presented that describes the tho-racic anatomy as a set of analytical primitives combined by means of Constructive Solid Geometry.

The matching of boundary templates is performed by either balancing an internal energy term with an external energy similar to snake-based approaches [144, 151], using global cost optimization strategies based on dynamic programming [144, 146-150], or by optimizing an energy function directly on explicit model parameters, thereby omit-ting an internal energy function [152-156]. During the matching, the allowed deforma-tion modes are restricted by prior knowledge, which is not necessarily populadeforma-tion-based. Such prior knowledge can be for instance an allowed shape interval on a set of radials along the template boundaries [145], assumptions about small deformations from the template shape [151], deformations along cascaded affine transforms [154-156], a restriction of the template deformations along orthogonal curves [144], or a restriction on the parameter bounds in analytical templates [153].

Due to the restricted degrees of deformation in boundary template matching, explicit prior knowledge can be imposed on the shape and its deformations resulting in more robust matching behavior, although boundary templates are less flexible than other snake approaches. Due to their relative rigidity, the boundary templates are applicable to single objects [144, 145] as well as to scenes consisting of multiple objects [151, 152, 154-156], where an optimum is sought for the scene model as a whole. Boundary templates can therefore be classified to the object level and the top (scene) level in Figure 2.1.

Volumetric templates

(21)

Since volumetric templates are matched as a whole, the topological structure of the model is preserved throughout the matching procedure, which makes it suitable to seg-ment multiple objects in a scene simultaneously. Therefore, these models can be placed on the scene interpretation level in the image interpretation pyramid in Figure 2.1.

High level scene models

On the highest level of abstraction in the image interpretation pyramid the high-level scene models can be found. These models explicitly describe knowledge about the scene domain under investigation. This can be knowledge about a typical size or volume of an organ, the typical image features for a set of organs in a particular image modality [168], the spatial relations of different organs with respect to each other [169-171] or the spa-tial embedding of objects in the image scene by means of a Voronoi diagram ([172], Chapter 3 [173] in this thesis). Two common representations for such high level knowl-edge are explicit rules [174, 175] and semantic- or frame networks [168-171, 176].

Explicit rules store knowledge in the form of ‘if ... then’ rules. This representation is often chosen to represent procedural knowledge consisting of a large number of discrete facts, and is highly flexible in its application, easy to extend and allows combination of multiple rules in a straighforward manner. Examples of the application of explicit rules to trace the intrathoracic airway trees in CT images and to semantically interpret MR brain scans are given in [175, 177] respectively.

Semantic- and frame networks are structured object descriptions, which describe a set of objects and their mutual relations as an attributed graph. Each object is described with a record data structure containing slots, which can be attributed a symbolic or numeric value. Relationships between different records are described by links, which characterize inheritance relations, neighborhood relations and part-subpart hierarchies. Frame- and semantic net representations are commonly applied in a framework for hypothesis gener-ation and verificgener-ation such as a blackboard system [168, 171]. In [170, 171, 175, 178], detailed implementations of frame representations have been described for knowledge driven segmentation of MR images of the brain, whereas applications to segmentation of thoracic- and abdominal CT scans have been described in [168, 169] respectively.

2.4 Summary

(22)

Anatomical models in medical image analysis 23 2.1). Anatomical models for segmentation purposes can be classified mainly to the high-est three levels in this image interpretation hierarchy.

Anatomical models on the lower image interpretation level are mainly limited in application to the formation of coherent scene elements such as contours and regions from low-level image information. Parametric and implicit snakes are examples of such models at this level, because the facilities to impose prior shape knowledge other than smoothness and connectivity constraints are limited. On the object recognition level, trainable models such as probabilistic snakes, point distribution models and object tem-plates can be distinguished, which combine prior knowledge about an average object shape and its characteristic shape deformations with domain specific low-level feature knowledge. At the highest level of abstraction, which involves the simultaneous process-ing of multiple objects within a scene, point-distribution models, volumetric template models and boundary template models can be utilized to describe the shapes of multiple organs in the scene within their spatial context. These models allow a coupling between high-level anatomical knowledge and low-level image features by defining different types of transforms, which enable a data driven deformation of the model to a feature pattern in the image data. A second class of anatomical models on the scene interpretation level are the high level scene models, which explicitly describe knowledge about the scene domain under investigation in a set of rules, a frame representation or a semantic net-work. These knowledge representations are commonly applied in a framework for hypothesis generation and verification such as a blackboard system.

2.5 References

[1] E. Richter, H. Krämer, W. Lierse, R. Maas, and K. H. Höhne, “Visualization of neonatal anatomy and pathology with a new computerized three-dimensional model as a basis for teaching, diagnosis and therapy,” Acta Anatomica, vol. 150, pp. 75-79, 1994.

[2] R. Schubert, K. H. Höhne, A. Pommert, M. Riemer, T. Schiemann, U. Tiede, and W. Lierse, “A new method for practicing exploration, dissection, and simulation with a complete computerized three-dimensional model of the brain and skull,” Acta Anatomica, vol. 150, pp. 69-74, 1994.

[3] K. H. Höhne, B. Pflesser, P. A., M. Riemer, T. Schiemann, R. Schubert, and U. Tiede, “A new repre-sentation of knowledge concerning anatomy and function,” Nature Medicine, vol. 1(6), pp. 506-511, 1995.

[4] T. Schiemann, J. Nuthmann, U. Tiede, and Höhne, “Generation of 3D anatomical atlases using the visible human,” in R. F. Kilcoyne, Proc. Computer Applications to Assist Radiology, pp. 62-67, Sym-posia foundation,1996.

[5] T. Schiemann, U. Tiede, and K. H. Höhne, “Segmentation and visualization of the Visible Human for high-quality volume-based visualization,” Medical Image Analysis, vol. 1(4), pp. 263-270, 1996. [6] V. Spitzer, M. J. Ackerman, A. L. Scherzinger, and D. Whitlock, “The Visible Human male: a

techni-cal report,” Journal of the American Meditechni-cal Informatics Association, vol. 3, pp. 118-130, 1996. [7] U. Tiede, T. Schiemann, and K. H. Höhne, “Visualizing the Visual Human,” IEEE Computer

(23)

[8] R. Mullick and Nguyen, “Visualization and labelling of the Visible Human dataset: challenges and resolves,” in K. H. Höhne and R. Kikinis, Proc. Visualization in Biomedical Computing, vol. 1131 of Lecture Notes in Computer Science, pp. 75-80, Springer Verlag, Berlin,1996.

[9] J. E. Stewart, W. C. Broaddus, and J. H. Johnson, “Rebuilding the Visual Man,” in K. H. Höhne and R. Kikinis, Proc. Visualization in Biomedical Computing, vol. 1131 of Lecture Notes in Computer Science, pp. 81-86, Springer Verlag, Berlin,1996.

[10] L. Serra, W. L. Nowinski, T. Poston, N. Hern, L. C. Meng, C. G. Guan, and P. K. Pillay, “The brain bench: virtual tools for stereotactic frame neurosurgery,” Medical Image Analysis, vol. 1(4), pp. 317-329, 1996.

[11] T. Schiemann and K. H. Höhne, “Definition of volume transformations for volume interaction,” in J. Duncan and G. Gindi, Proc. Information Processing in Medical Imaging, vol. 1230 of Lecture Notes in Computer Science, pp. 245-258, Springer Verlag, Berlin,1997.

[12] S. Gibson, C. Fyock, E. Grimson, T. Kanade, R. Kikinis, H. Lauer, N. McKenzie, A. Mor, S. Naka-jima, H. Ohkami, R. Osborne, J. Samosky, and A. Sawada, “Volumetric object modeling for surgical simulation,” Medical Image Analysis, vol. 2(2), pp. 121-132, 1998.

[13] H. Fuchs, A. State, E. D. Pisano, W. F. Garret, G. Hirota, M. Livingston, M. C. Whitton, and S. M. Pizer, “Towards performing ultrasound-guided needle biopsies from within a head-mounted display,” in K. H. Höhne and R. Kikinis, Proc. Visualization in Biomedical Computing, vol. 1131 of Lecture Notes in Computer Science, pp. 591-600, Springer Verlag, Berlin,1996.

[14] Y. Sato, M. Nakamoto, Y. Tamaki, T. Sasama, I. Sakita, Y.Nakajima, M. Monden, and S. Tamura, “Image guidance of breast cancer surgery using 3-D ultrasound images and augmented reality visual-ization,” IEEE Transactions on Medical Imaging, vol. 17(5), pp. 681-693, 1998.

[15] W. M. Wells, A. C. F. Colchester, and S. Delp, Proc. “Medical Image Computing and Computer Assisted Intervention,” Lecture Notes in Computer Science, vol. 1496: Springer Verlag, Berlin, 1998, pp. 1258.

[16] I. G. Zubal and C. R. Harrell, “Voxel based Monte-Carlo calculations of nuclear medicine images and applied variance reduction techniques,” Image and Vision Computing, vol. 10, pp. 342-348, 1992. [17] I. G. Zubal, C. R. Harrell, E. O. Smith, Z. Rattner, G. G., and P. B. Hoffer, “Computerized

three-dimensional segmented human anatomy,” Medical Physics, vol. 21(2), pp. 299-302, 1994. [18] H. Wang, R. J. Jaszczak, and R. E. Coleman, “Solid geometry-based object model for Monte Carlo

simulated emission and transmission tomographic imaging systems,” IEEE Transactions on Medical

Imaging, vol. 11(3), pp. 361-372, 1992.

[19] C.-L. Huang, W.-T. Chang, L.-C. Wu, and J.-K. Wang, “Three-dimensional PET emission scan reg-istration and transmission scan synthesis,” IEEE Transactions on Medical Imaging, vol. 16(6), 1997. [20] R. K. S. Kwan, A. C. Evans, and G. B. Pike, “An extensible MRI simulator for post-processing

evalu-ation,” in K. H. Höhne and R. Kikinis, Proc. Visualization in Biomedical Computing, vol. 1131 of Lecture Notes in Computer Science, pp. 135-140, Springer Verlag, Berlin,1996.

[21] D. L. Collins, A. P. Zijdenbos, V. Kollokian, J. G. Sled, N. J. Kabani, C. J. Holmes, and A. C. Evans, “Design and construction of a realistic digital brain phantom,” IEEE Transactions On Medical

Imag-ing, vol. 17(3), pp. 463-468, 1998.

(24)

Anatomical models in medical image analysis 25 [23] Z. A. Cohen, D. M. McCarthy, H. Roglic, J. H. Henry, W. G. Rodkey, J. R. Steadman, V.C. Mow,

and G. A. Ateshian, “Computer-aided planning of patellofemoral joint OA surgery: developing phys-ical models from patient MRI,” in W. M. Wells and A. C. F. Colchester, Proc. MICCAI, vol. 1496 of Lecture Notes in Computer Science, pp. 9-20, Springer Verlag, Berlin,1998.

[24] P. Edwards, D. Hill, J. Little, and D. Hawkes, “A three-component deformation model for image-guided surgery,” Medical Image Analysis, vol. 2(4), pp. 355-367, 1998.

[25] Y. Zhu, M. Drangova, and N. J. Pelc, “Estimation of deformation gradient and strain from cine-PC velocity data,” IEEE Transactions on Medical Imaging, vol. 16(6), pp. 840-851, 1997.

[26] J. Declerck, J. Feldmar, M. Goris, and F. Betting, “Automatic registration and alignment on a tem-plate of cardiac stress and rest reoriented SPECT images,” IEEE Transactions on Medical Imaging, vol. 16(6), pp. 727-737, 1997.

[27] J. Declerck, J. Feldmar, and N. Ayache, “Definition of a four-dimensional continuous planispheric transformation for the tracking and the analysis of left-ventricle motion,” Medical Image Analysis, vol. 2(2), pp. 197-213, 1998.

[28] S. Benayoun, C. Nastar, and N. Ayache, “Dense non-rigid motion estimation in sequences of 3D images using differential constraints,” in N. Ayache, Proc. CVRMed, vol. 905 of Lecture Notes in Computer Science, pp. 309-318, Springer Verlag, Berlin,1995.

[29] E. Bardinet, L. D. Cohen, and N. Ayache, “Tracking and motion analysis of the left ventricle with deformable superquadrics,” Medical Image Analysis, vol. 1(2), pp. 129-149, 1996.

[30] P. Clarysse, D. Friboulet, and I. E. Magnin, “Tracking geometrical descriptors on 3-D deformable surfaces: application to the left-ventricular surface of the heart,” IEEE Transactions on Medical

Imag-ing, vol. 16(4), pp. 392-404, 1997.

[31] J. L. Prince and E. R. McVeigh, “Motion estimation from tagged MR image sequences,” IEEE

Trans-actions on Medical Imaging, vol. 11(2), pp. 238-249, 1992.

[32] A. A. Young and L. Axel, “Tracking and finite element analysis of stripe deformation in magnetic res-onance tagging,” IEEE Transactions on Medical Imaging, vol. 14(3), pp. 413-421, 1995.

[33] D. Friboulet, I. E. Magnin, and D. Revel, “Assessment of a model for overall left ventricular three-dimensional motion from MRI-data,” The International Journal of Cardiac Imaging, vol. 8, pp. 175-190, 1992.

[34] J. M. Gorce, D. Friboulet, P. Clarysse, and I. E. Magnin, “Three-dimensional velocity field estima-tion of moving cardiac walls,” in Proc. Computers in Cardiology, pp. 489-492, 1994.

[35] W. C. Huang and D. B. Goldgof, “Adaptive-size meshes for rigid and nonrigid shape analysis and synthesis,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 15(6), pp. 611-616, 1993.

[36] C. W. Chen, T. S. Huang, and M. Arrot, “Modeling, analysis and visualization of left ventricle shape and motion by hierarchical decomposition,” IEEE Transactions on Pattern Analysis and Machine

Intel-ligence, vol. 16(4), pp. 324-356, 1994.

[37] J. Duncan, R. Owen, P. Anandan, L. Staib, T. McCauley, A. Salazar, and F. Lee, “Shape-based track-ing of left ventricular wall motion,” in Proc. Computers in Cardiology, pp. 41-44, 1991.

(25)

[39] J. S. Duncan, F. A. Lee, A. W. M. Smeulders, and B. L. Zaret, “A bending energy model for measure-ment of cardiac shape deformity,” IEEE Transactions on Medical Imaging, vol. 10(3), pp. 307-320, 1991.

[40] W. G. O'Dell, C. C. Moore, W. C. Hunter, E. A. Zerhouni, and E. R. McVeigh, “Three-dimensional myocardial deformations: calculation with displacement field fitting to tagged MR-images,”

Radiol-ogy, vol. 195(3), pp. 165-175, 1995.

[41] H. Azhari, S. Sideman, R. Beyar, E. Grenadier, and U. Dinnar, “An analytical shape descriptor of 3-D geometry. Application to the analysis of the left ventricular shape and contraction,” IEEE Transactions

on Biomedical Engineering, 1987.

[42] J. Park, D. Metaxas, and L. Axel, “Analysis of left ventricular wall motion based on volumetric deformable models and MRI-SPAMM,” Medical Image Analysis, vol. 1(1), pp. 53-71, 1996. [43] J. Park, D. Metaxas, A. A. Young, and L. Axel, “Deformable models with parameter functions for

car-diac motion analysis from tagged MRI data,” IEEE Transactions on Medical Imaging, vol. 15(3), pp. 278-289, 1996.

[44] E. L. Dove, K. P. Philip, D. D. McPherson, and B. Chandran, “Quantitative shape descriptors of left ventricular cine-CT images,” IEEE Transactions on Biomedical Engineering, vol. 38(12), pp. 1256-1261, 1991.

[45] S. K. Mishra, D. B. Goldgof, and T. S. Huang, “Motion analysis and epicardial deformation estima-tion from angiographic data,” in Proc. Computer Vision and Pattern Recogniestima-tion, pp. 331-336, 1991.

[46] A. Pentland and B. Horowitz, “Recovery of nonrigid motion and structure,” IEEE Transactions on

Pattern Analysis and Machine Intelligence, vol. 13(7), pp. 730-742, 1991.

[47] J. C. McEachen and J. S. Duncan, “Shape-based tracking of left ventricular wall motion,” IEEE

Transactions on Medical Imaging, vol. 16(3), pp. 270-283, 1997.

[48] F. G. Meyer, R. T. Constable, A. J. Sinusas, and J. S. Duncan, “Tracking myocardial deformation using phase constrast MR velocity fields: a stochastic approach,” IEEE Transactions on Medical

Imag-ing, vol. 15(4), pp. 453-465, 1996.

[49] C. Nastar and N. Ayache, “A new physically based model for efficient tracking and analysis of defor-mations,” Lecture Notes in Computer Science, vol. 911, pp. 239-283, 1993.

[50] P. Shi, G. Robinson, A. Chakraborty, L. Staib, R. Constable, A. Sinusas, and J. Duncan, “A unified framework to assess myocardial function from 4D images,” Lecture Notes in Computer Science, vol. 905, pp. 327-340, 1995.

[51] C. Nastar and N. Ayache, “Non-rigid motion analysis in medical images: A physically based approach,” Lecture Notes in Computer Science, vol. 687, pp. 17-32, 1993.

[52] C. Nastar, “Vibration modes for nonrigid motion analysis in 3D images,” Lecture Notes in Computer

Science, vol. 801, pp. 231-236, 1994.

[53] A. Gupta, T. O'Donnel, and A. Singh, “Segmentation and tracking of cine cardiac MR and CT images using a 3-D deformable model,” in Proc. Computers in Cardiology, pp. 661-664, 1994. [54] P. Radeva, A. A. Amini, and J. T. Huang, “Deformable B-solids and implicit snakes for 3D

localiza-tion and tracking of SPAMM MRI data,” Computer Vision and Image Understanding, vol. 66(2), pp. 163-178, 1997.

(26)

Anatomical models in medical image analysis 27 [56] R. T. Constable, K. M. Rath, A. J. Sinusas, and G. G. J, “Development and evaluation of tracking

algorithms for cardiac wall motion analysis using phase velocity MR imaging,” Magnetic Resonance in

Medicine, vol. 32, pp. 33-42, 1994.

[57] A. A. Young, “Model Tags: Direct 3D tracking of heart wall motion from tagged MR images,” in W. M. Wells and A. C. F. Colchester, Proc. MICCAI, vol. 1496 of Lecture Notes in Computer Science, pp. 92-101, Springer Verlag, Berlin,1998.

[58] A. A. Young, D. L. Kraitchman, and L. Axel, “Deformable models for tagged MR images: reconstruc-tion of two- and three-dimensional heart wall moreconstruc-tion,” in Proc. IEEE Workshop on Biomedical image analysis, pp. 317-332, 1994.

[59] H. D. Tagare, “Non-rigid curve correspondence for estimating heart motion,” in J. Duncan and G. Gindi, Proc. Information Processing in Medical Imaging, vol. 1230 of Lecture Notes in Computer Science, pp. 489-494, Springer Verlag, Berlin,1997.

[60] S. Sullivan, L. Sandford, and J. Ponce, “Using geometric distance fits for 3-D object modeling and recognition,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 16(12), pp. 1183-1195, 1994.

[61] T. O'Donnel, A. Gupta, and T. Boult, “The hybrid volume ventriculoid: a model for MR-SPAMM 3-D analysis,” in Proc. Computers in Cardiology, pp. 5-8, 1995.

[62] C. Nastar and N. Ayache, “Frequency-based nonrigid motion analysis: application to four dimen-sional medical images,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 18(11), 1996.

[63] E. Haber, D. Metaxas, and L. Axel, “Motion analysis of the right ventricle from MRI images,” in W. M. Wells and A. C. F. Colchester, in Proc. MICCAI, vol. 1496 of Lecture Notes in Computer Sci-ence, pp. 177-188, Springer Verlag, Berlin,1998.

[64] G. Subsol, J. P. Thirion, and N. Ayache, “A scheme for automatically building three-dimensional morphometric anatomical atlases: application to a skull atlas,” Medical Image Analysis, vol. 2(1), pp. 37-60, 1998.

[65] F. Bookstein, “Landmark methods for forms without landmarks: morphometrics of group differences in outline shape,” Medical Image Analysis, vol. 1(3), pp. 225-243, 1996.

[66] F. Bookstein, “Visualizing group differences in outline shape: methods from biometrics of landmark points,” in K. H. Höhne and R. Kikinis, in Proc. Visualization in Biomedical Computing, vol. 1131 of Lecture Notes in Computer Science, pp. 405-410, Springer Verlag, Berlin,1996.

[67] F. L. Bookstein, “Combining "vertical" and "horizontal" features from medical images,” in Lecture

Notes in Computer Science, vol. 905, 1995, pp. 184-191.

[68] F. L. Bookstein, “Principal warps: thin-plate splines and the decomposition of deformations,” IEEE

Transactions on Pattern Analysis and Machine Intelligence, vol. 11(6), pp. 567-585, 1989.

[69] F. L. Bookstein, “Quadratic variation of deformations,” in J. Duncan and G. Gindi, Proc. Informa-tion Processing in Medical Imaging, vol. 1230 of Lecture Notes in Computer Science, Springer Ver-lag, Berlin,1997.

[70] C. Davatzikos, “Spatial normalization of 3D brain images using deformable models,” Journal of

Com-puter Assisted Tomography, vol. 20(4), pp. 656-665, 1996.

[71] G. E. Christensen, S. C. Joshi, and M. I. Miller, “Volumetric transformation of brain anatomy,” IEEE

(27)

[72] S. Joshi, A. Banerjee, G. E. Christensen, J. G. Csernansky, J. W. Haller, M. I. Miller, and L. Wang, “Gaussian random fields on sub-manifolds for characterizing brain surfaces,” in J. Duncan and G. Gindi, Proc. Information Processing in Medical Imaging, vol. 1230 of Lecture Notes in Computer Science, Springer Verlag, Berlin,1997.

[73] M. Miller, A. Banerjee, G. Christensen, S. Joshi, N. Khaneja, U. Grenander, and L. Matejic, “Statisti-cal methods in computational anatomy,” Statisti“Statisti-cal Methods in Biomedi“Statisti-cal Research, vol. 6, pp. 267-299, 1997.

[74] P. Thompson and A. W. Toga, “Visualization and mapping of anatomic abnormalities using a proba-bilistic brain atlas based on random fluid transformations.,” in K. H. Höhne and R. Kikinis, Proc. Visualization in Biomedical Computing, vol. 1131 of Lecture Notes in Computer Science, pp. 383-392, Springer Verlag, Berlin,1996.

[75] D. Dean, P. Buckley, F. Bookstein, J. Kamath, D. Kwon, L. Friedman, and C. Lys, “Three-dimen-sional MR based, morphometric comparison of schizophrenic and normal cerebral ventricles,” in K. H. Höhne and R. Kikinis, Proc. Visualization in Biomedical Computing, vol. 1131 of Lecture Notes in Computer Science, pp. 363-372, Springer Verlag, Berlin,1996.

[76] P. M. Thompson and A. W. Toga, “Detection, visualization and animation of abnormal anatomic structure with a deformable probabilistic brain atlas based on random vector field transformations,”

Medical Image Analysis, vol. 1(4), pp. 271-294, 1996.

[77] A. Singh, D. Goldgof, and D. Terzopoulos, “Deformable models in medical image analysis,” . Los Alamitos, CA: IEEE Computer Society Press, 1998.

[78] D. Terzopoulos, J. Platt, A. Barr, and K. Fleischer, “Elastically deformable models,” Comp. Graph, vol. 21(4), pp. 205-214, 1987.

[79] M. Kass, A. Witkin, and D. Terzopoulos, “Snakes: Active contour models,” International Journal of

Computer Vision, vol. 1(4), pp. 321-331, 1988.

[80] J. V. Miller, D. E. Breen, W. E. Lorensen, R. M. O'Bara, and M. J. Wozny, “Geometrically deformed models: a method for extracting closed geometric models from volume data,” Computer Graphics, vol. 25(4), pp. 217-226, 1991.

[81] L. D. Cohen, “On active contour models and balloons,” CVGIP: Image Understanding, vol. 53(2), pp. 211-218, 1991.

[82] I. Cohen, L. D. Cohen, and N. Ayache, “Using deformable surfaces to segment 3-D images and infer differential structures,” CVGIP: Image Understanding, vol. 56(2), pp. 242-263, 1992.

[83] L. D. Cohen and I. Cohen, “Finite-element methods for active contour models and balloons for 2-D and 3-D images,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 15(11), pp. 1131-1147, 1993.

[84] T. McInerney and D. Terzopoulos, “A dynamic finite element surface model for segmentation and tracking in multidimensional medical images with application to cardiac 4D image analysis,”

Com-puterized Medical Imaging and Graphics, vol. 19, pp. 69-83, 1995.

[85] T. N. Jones, “Automated 3D segmentation using deformable models and fuzzy affinity,” in J. Duncan and G. Gindi, Proc. Information Processing in Medical Imaging, vol. 1230 of Lecture Notes in Com-puter Science, pp. 113-126, Springer Verlag, Berlin,1997.

(28)

Anatomical models in medical image analysis 29 [87] L. Gao, D. Heath, and E. K. Fishman, “Abdominal Image Segmentation Using Three-Dimensional

Deformable Models,” Journal of Computer Assisted Tomography, vol. 33(6), pp. 348-355, 1998. [88] H. Delingette, M. Hebert, and K. Ikeuchi, “Shape representation and image segmentation using

deformable surfaces,” Image and Vision Computing, vol. 10(3), pp. 132-144, 1992.

[89] C. Davatzikos and R. N. Bryan, “Using a deformable surface model to obtain a shape representation of the cortex,” IEEE Transactions on Medical Imaging, vol. 15(6), pp. 785-795, 1996.

[90] C. Davatzikos, “Spatial transformation and registration of brain images using elastically deformable models,” Computer Vision and Image Understanding, vol. 66 (2), pp. 207-222, 1997.

[91] S. Sandor and R. Leahy, “Surface-based labeling of cortical anatomy using a deformable atlas,” IEEE

[92] M. Vaillant and C. Davatzikos, “Finding parametric representations of the cortical sulci using an active contour model,” Medical Image Analysis, vol. 1(4), pp. 295-315, 1996.

[93] D. Terzopoulos and D. Metaxas, “Dynamic 3D models with local and global deformations: deform-able superquadrics,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 13(7), pp. 703-714, 1991.

[94] J. G. Snel, H. W. Venema, and C. A. Grimbergen, “Detection of the Carpal Bone Contours from 3-D MR Images of the Wrist Using a Planar Radial Scale-Space Snake,” IEEE Transactions on Medical

Imaging, vol. 17(6), pp. 1049-1062, 1999.

[95] A. Chakraborty, L. H. Staib, and J. S. Duncan, “Deformable boundary finding in medical images by integrating gradient and region information,” IEEE Transactions on Medical Imaging, vol. 15(6), pp. 859-870, 1996.

[96] M. Worring, A. W. M. Smeulders, L. H. Staib, and J. S. Duncan, “Parametrized feasible boundaries in gradient vector fields,” Computer Vision and Image Understanding, vol. 63(1), pp. 135-144, 1996. [97] F. Leitner and P. Cinquin, “From splines and snakes to SNAKE SPLINES,” Lecture Notes in Computer

Science, vol. 911, pp. 264-281, 1991.

[98] S. Lobregt and M. A. Viergever, “A discrete dynamic contour model,” IEEE Transactions on Medical

Imaging, vol. 14(1), pp. 12-24, 1995.

[99] T. B. Sebastian, H. Tek, J. J. Crisco, S. W. Wolfe, and B. B. Kimia, “Segmentation of carpal bones from 3D CT images using skeletally coupled deformable models,” in W. M. Wells and A. C. F. Colchester, Proc. MICCAI, vol. 1496 of Lecture Notes in Computer Science, pp. 1185-1194, Springer Verlag, Berlin,1998.

[100] A. A. Amini, T. E. Weymouth, and R. C. Jain, “Using dynamic programming for solving variational problems in vision,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 12(9), pp. 855-867, 1990.

[101] S. Ranganath, “Contour Extraction from Cardiac MRI Studies Using Snakes,” IEEE Transactions on

Medical Imaging, vol. 14(2), pp. 328-338, 1995.

[102] A. Goshtasby and D. A. Turner, “Segmentation of Cardiac Cine MR Images for Extraction of Right and Left Ventricular Chambers,” IEEE Transactions on Medical Imaging, vol. 14(1), pp. 56-64, 1995. [103] L. H. Staib and J. S. Duncan, “Model-based deformable surface finding for medical images,” IEEE

(29)

[105] V. Cassales, F. Catte, T. Coll, and F. Dibos, “A geometric model for active contours in image process-ing,” Numerische Mathematik, vol. 66, pp. 1-31, 1993.

[106] R. Malladi, J. A. Sethian, and B. C. Vemuri, “Shape modeling with front propagation: a level set approach,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 17(2), pp. 158-175, 1995.

[107] T. McInerney and D. Terzopoulos, “Medical image segmentation using topologically adaptable snakes,” Lecture Notes in Computer Science, vol. 905, pp. 92-104, 1995.

[108] V. Casalles, R. Kimmel, and G. Sapiro, “Geodesic active contours,” International Journal of Computer

Vision, vol. 22(1), pp. 61-79, 1997.

[109] A. Yezzi, S. Kichenassamy, A. Kumar, P. Olver, and A. Tannenbaum, “A geometric snake model for segmentation of medical imagery,” IEEE Transactions on Medical Imaging, vol. 16(2), pp. 199-209, 1997.

[110] L. M. Lorigo, O. Faugeras, W. E. L. Grimson, R. Keriven, and R. Kikinis, “Segmentation of bone in clinical knee MRI using texture-based geodesic active contours,” in Proc. MICCAI, Lecture Notes in Computer Science, vol. 1496, pp. 1195-1204, 1998.

[111] W. J. Niessen, B. M. ter Haar Romeny, and M. A. Viergever, “Geodesic deformable models for medi-cal image analysis,” IEEE Transactions on Medimedi-cal Imaging, vol. 17(4), pp. 634-641, 1998.

[112] X. Zheng, L. H. Staib, R. T. Schultz, and J. S. Duncan, “Segmentation and measurement of the cor-tex from 3D MR images,” in Proc. MICCAI, Lecture Notes in Computer Science, vol. 1496, pp. 519-530, 1998.

[113] B. C. Vemuri, A. Radisavljevic, and C. M. Leonard, “Multi-resolution Stochastic 3D Shape Models for Image Segmentation,” Lecture Notes in Computer Science, vol. 687, pp. 62-76, 1993.

[114] B. C. Vemuri and A. Radisavljevic, “Multiresolution stochastic hybrid shape models with fractal pri-ors,” ACM Transactions on Graphics, vol. 13(2), pp. 177-207, 1994.

[115] B. C. Vemuri, Y. Guo, C. M. Leonard, and S.-H. Lai, “Fast numerical algorithms for fitting multires-olution hybrid shape models to brain MRI,” Medical Image Analysis, vol. 1(4), pp. 343-362, 1996. [116] L. H. Staib and J. S. Duncan, “Boundary finding with parametrically deformable contour models,”

IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 14(11), pp. 1061-1075, 1992.

[117] C. Brechbühler, G. Gerig, and O. Kubler, “Parametrization of closed surfaces for 3-D shape descrip-tion,” Computer Vision and Image Understanding, vol. 61(2), pp. 154-170, 1995.

[118] G. Szekely, A. Kelemen, C. Brechbühler, and G. Gerig, “Segmentation of 3D objects from MRI vol-ume data using constrained elastic deformation of flexible Fourier surface models,” Lecture Notes in

Computer Science, vol. 905, pp. 494-505, 1995.

[119] G. Szekely, A. Kelemen, C. Brechbühler, and G. Gerig, “Segmentation of 2-D and 3-D objects from MRI volume data using constrained elastic deformations of flexible Fourier contour and surface mod-els,” Medical Image Analysis, vol. 1(1), pp. 19-34, 1996.

[120] T. F. Cootes, A. Hill, C. J. Taylor, and J. Haslam, “The use of active shape models for locating struc-tures in medical images,” Lecture Notes in Computer Science, vol. 687, pp. 33-47, 1993.

[121] T. F. Cootes, A. Hill, C. J. Taylor, and J. Haslam, “Use of active shape models for locating structures in medical images,” Image and Vision Computing, vol. 12(6), pp. 355-366, 1994.

(30)

Anatomical models in medical image analysis 31 [123] A. Hill, T. F. Cootes, and C. J. Taylor, “A genetic system for image interpretation using flexible

tem-plates,” in Proc. British Machine Vision Conference, 1992.

[124] A. Hill and C. J. Taylor, “Model based image interpretation using genetic algorithms,” Image and

Vision Computing, vol. 10, pp. 295-300, 1992.

[125] A. Hill, T. F. Cootes, C. J. Taylor, and K. Lindley, “Medical image interpretation: a generic approach using deformable templates,” Medical Informatics, vol. 19(1), pp. 47-60, 1994.

[126] A. C. W. Kotcheff and C. J. Taylor, “Automatic reconstruction of eigenshape models by genetic algo-rithm,” in J. Duncan and G. Gindi, Proc. Information Processing in Medical Imaging, vol. 1230 of Lecture Notes in Computer Science, pp. 441-446, Springer Verlag, Berlin,1997.

[127] A. C. W. Kotcheff and C. J. Taylor, “Automatic construction of eigenshape models by direct optimi-zation,” Medical Image Analysis, vol. 2(4), pp. 303-314, 1998.

[128] S. Sclaroff and A. P. Pentland, “Modal matching for correspondence and recognition,” IEEE

Transac-tions on Pattern Analysis and Machine Intelligence, vol. 17(6), pp. 545-561, 1995.

[129] C. Kambhamettu and D. B. Goldgof, “Curvature-based approach to point correspondence recovery in conformal nonrigid motion,” CVGIP: Image Understanding, vol. 60(1), pp. 26-43, 1994. [130] A. Hill, A. D. Brett, and C. J. Taylor, “Automatic landmark identification using a new method of

non-rigid correspondence,” in J. Duncan and G. Gindi, Proc. Information Processing in Medical Imaging, vol. 1230 of Lecture Notes in Computer Science, pp. 483-488, Springer Verlag, Ber-lin,1997.

[131] N. Duta and M. Sonka, “Segmentation and interpretation of MR brain images using an improved knowledge-based active shape model,” in J. Duncan and G. Gindi, Proc. Information Processing in Medical Imaging, vol. 1230 of Lecture Notes in Computer Science, pp. 381-386, Springer Verlag, Berlin,1997.

[132] P. P. Smyth, C. J. Taylor, and J. E. Adams, “Automatic measurement of vertebral shape using active shape models,” in J. Duncan and G. Gindi, Proc. Information Processing in Medical Imaging, vol. 1230 of Lecture Notes in Computer Science, pp. 441-446, Springer Verlag, Berlin,1997.

[133] N. Duta and M. Sonka, “Segmentation and interpretation of MR brain images: an improved active shape model,” IEEE Transactions on Medical Imaging, vol. 17(6), pp. 1049-1062, 1999.

[134] T. L. Faber, E. M. Stokely, R. M. Peshock, and J. R. Corbett, “A model-based four-dimensional left ventricular surface detector,” IEEE Transactions on Medical Imaging, vol. 10(3), pp. 321-329, 1991. [135] A. Zijdenbos, A. C. Evans, F. Riahi, J. Sled, J. Chui, and V. Kollokian, “Automatic quantification of

multiple sclerosis lesion volume using stereotaxic space,” in K. H. Höhne and R. Kikinis, Proc. Visu-alization in Biomedical Computing, vol. 1131 of Lecture Notes in Computer Science, pp. 439-448, Springer Verlag, Berlin,1996.

[136] D. L. Collins, P. Neelin, T. M. Peters, and A. C. Evans, “Automatic 3D intersubject registration of MR volumetric data in standardized Talairach space,” Journal of Computer Assisted Tomography, vol. 18(2), pp. 192-205, 1994.

[137] D. L. Collins, C. J. Holmes, T. M. Peters, and A. C. Evans, “Automatic 3-D model-based neuroana-tomical segmentation,” Human Brain Mapping, vol. 3(3), pp. 190-208, 1995.

[138] N. Karssemeijer, “A statistical method for automatic labeling of tissues in medical images,” Machine

(31)

[139] K. van Leemput, F. Maes, D. Vandermeulen, and P. Suetens, “Automatic segmentation of brain tissues and MR bias field correction using a digital brain atlas,” in W. M. Wells and A. C. F. Colchester, Proc. MICCAI, vol. 1496 of Lecture Notes in Computer Science, pp. 1222-1229, Springer Verlag, Ber-lin,1998.

[140] G. Le Goualher, D. L. Collins, C. Barillot, and A. C. Evans, “Automatic identification of cortical sulci using a 3D probabilistic atlas,” in W. M. Wells and A. C. F. Colchester, Proc. MICCAI, vol. 1496 of Lecture Notes in Computer Science, pp. 509-518, Springer Verlag, Berlin,1998.

[141] J. L. Boes, P. H. Bland, T. E. Weymouth, L. E. Quint, F. L. Bookstein, and C. R. Meyer, “Generating a normalized geometric liver model using warping,” Investigative Radiology, vol. 29(3), pp. 281-286, 1994.

[142] J. L. Boes, C. Meyer, and T. E. Weymouth, “Liver definition in CT using a population-based shape model,” Lecture Notes in Computer Science, vol. 905, pp. 506-512, 1995.

[143] J. L. Boes, T. E. Weymouth, and C. R. Meyer, “Multiple organ definition in CT using a Bayesian approach for 3D model fitting,” Proc. SPIE, vol. 2573, pp. 244-251, 1995.

[144] H. D. Tagare, “Deformable 2-D template matching using orthogonal curves,” IEEE Transactions on

Medical Imaging, vol. 16(1), pp. 108-117, 1997.

[145] J. Brinkley, “A flexible, generic model for anatomic shape: application to interactive two-dimensional medical image segmentation and matching,” Computers and Biomedical Research, vol. 26, pp. 121-142, 1993.

[146] J. G. Bosch, L. H. Savalle, G. van Burken, and J. H. C. Reiber, “Evaluation of a semiautomatic con-tour detection approach in sequences of short-axis two-dimensional echocardiographic images,”

Jour-nal of the American Society Echocardiography, vol. 8, pp. 810-821, 1995.

[147] R. J. van der Geest, V. G. M. Buller, E. Jansen, H. J. Lamb, L. H. B. Baur, E. E. van der Wall, A. de Roos, and J. H. C. Reiber, “Comparison between manual and semiautomated analysis of left ventric-ular volume parameters from short-axis MR images,” Journal of Computer Assisted Tomography, vol. 21(5), pp. 756-765, 1997.

[148] R. J. van der Geest, R. A. Niezen, E. E. van der Wall, A. de Roos, and J. H. C. Reiber, “Automated measurement of volume flow in the ascending aorta using MR velocity maps: evaluation of inter- and interobserver variability in healthy volunteers,” Journal of Computer Assisted Tomography, vol. 22(6), pp. 904-911, 1998.

[149] M. Sonka, X. Zhang, M. Siebes, M. S. Bissing, S. DeJong, S. M. Collins, and C. R. McKay, “Seg-mentation of intravascular ultrasound images: A knowledge guided approach.,” IEEE Transactions on

Medical Imaging, vol. 14, pp. 719-732, 1995.

[150] A. Krivanek and M. Sonka, “Ovarian ultrasound image analysis: follicle segmentation,” IEEE

Trans-actions on Medical Imaging, vol. 17(6), pp. 935-944, 1998.

[151] J. Lötjönen, I. E. Magnin, P.-J. Reissman, J. Nenonen, and T. Katila, “Segmentation of magnetic res-onance images using 3D deformable models,” in proc. MICCAI, Lecture Notes in Computer Sci-ence, vol. 1496, pp. 9-20, 1998.

[152] A. L. Yuille, P. W. Hallinan, and D. S. Cohen, “Feature extraction from faces using deformable tem-plates,” International Journal of Computer Vision, vol. 8(2), pp. 99-111, 1992.

[153] K. Delibasis and P. E. Undrill, “Anatomical object recognition using deformable geometric models,”

(32)

Anatomical models in medical image analysis 33 [154] B. P. F. Lelieveldt, M. Sonka, L. Bolinger, T. D. Scholtz, H. W. M. Kayser, R. J. v. d. Geest, and J. H. C. Reiber, “Anatomical modeling with fuzzy implicit surfaces: application to automated localization of the heart and lung surfaces in thoracic MR Images,” in A. Kuba and M. Samal, Proc. Information Processing in Medical Imaging, vol. 1613 of Lecture Notes in Computer Science, pp 400-405, Springer Verlag, Berlin,1999.

[155] B. P. F. Lelieveldt, R. J. van der Geest, and J. H. C. Reiber, “Automated model driven localization of the heart and lung surfaces in thoracic MR images,” Computers in Cardiology, vol. 25, pp. 9-12, 1998. [156] B. P. F. Lelieveldt, R. J. van der Geest, M. Ramze Rezaee, J. G. Bosch, and J. H. C. Reiber, “Anatom-ical model matching with fuzzy implicit surfaces for segmentation of thoracic volume scans,” IEEE

Transactions on Medical Imaging, vol. 18(2), pp 218-230, 1999.

[157] P. St-Jean, A. F. Sadikot, L. Collins, D. Clonda, R. Kasrai, A. C. Evans, and T. M. Peters, “Automated atlas integration and interactive three-dimensional vsualization tools for planning and guidance in functional neurosurgery,” IEEE Transactions on Medical Imaging, vol. 17(5), pp. 672-680, 1998. [158] T. Greitz, C. Bohm, Holte.S., and Eriksson, “A computerized brain atlas: construction, anatomical

content and some applications,” Journal of Computer Assisted Tomography, vol. 15, pp. 26-38, 1991. [159] R. Dann, J. Hoford, S. Kovacic, M. Reivich, and R. Bajcsy, “Evaluation of elastic matching system for

anatomic (CT,MR) and functional (PET) cerebral images,” Journal of Computed Tomography, vol. 13(4), pp. 603-611, 1989.

[160] R. Bajcsy and S. Kovacic, “Multiresolution elastic matching,” Computer Vision, Graphics and Image

Processing, vol. 46, pp. 1-21, 1989.

[161] A. C. Evans, W. Dai, L. Collins, P. Neelin, and S. Marret, “Warping of a computerized 3-D atlas to match brain image volumes for quantitative neuroanatomical and functional analysis,” Proceedings

SPIE Image Processing, vol. 1445, pp. 236-247, 1991.

[162] G. E. Christensen, R. D. Rabbitt, and M. I. Miller, “3D-brain mapping using a deformable neu-roanatomy,” Physics in Medicine and Biology, vol. 39, pp. 609-618, 1994.

[163] G. E. Christensen, R. D. Rabbitt, and M. I. Miller, “Deformable templates using large deformation kinematics,” IEEE Transactions on Image Processing, vol. 5(10), pp. 1435-1447, 1996.

[164] J. W. Haller, A. Banerjee, G. E. Christensen, M. Gado, S. Joshi, M. I. Miller, Y. Sheline, M. W. Van-nier, and J. G. Csernansky, “Three-dimensional hippocampal MR morphometry with high-dimen-sional transformation of a neuroanatomic atlas,” Radiology, vol. 202(2), pp. 504-510, 1997. [165] M. Bro-Nielsen and C. Gramkow, “Fast fluid registration of medical images,” in K. H. Höhne and R.

Kikinis, Proc. Visualization in Biomedical Computing, vol. 1131 of Lecture Notes in Computer Sci-ence, pp. 267-276, Springer Verlag, Berlin,1996.

[166] J. P. Thirion, “Image matching as a diffusion process: an analogy with Maxwell's demons,” Medical

Image Analysis, vol. 2(3), pp. 243-260, 1998.

[167] Y. Wang and L. Staib, “Elastic model based non-rigid registration incorporating statistical shape information,” in W. M. Wells and A. C. F. Colchester, Proc. MICCAI, vol. 1496 of Lecture Notes in Computer Science, pp. 1162-1173, Springer Verlag, Berlin,1998.

(33)

[169] N. Karssemeijer, L. J. Erning, O. v. a. n. Th, and E. G. J. Eikman, “Recognition of organs in CT-image sequences: a model guided approach,” Computers and Biomedical Research, vol. 21(5), pp. 434-448, 1988.

[170] G. P. Robinson, A. C. F. Colchester, and L. D. Griffin, “Model-based recognition of anatomical objects from medical images,” Lecture Notes in Computer Science, vol. 687, pp. 197-211, 1993. [171] H. Li, R. Deklerck, De, B. Cuyper, A. Hermanus, E. Nyssen, and J. Cornelis, “Object Recognition in

Brain CT Scans: Knowledge-Based Fusion of Data From Multiple Feature Extractors,” IEEE

Transac-tions on Medical Imaging, vol. 14(2), pp. 212-229, 1995.

[172] F. Poupon, J.-F. Magnin, D. Hasboun, C. Poupon, I. Magnin, and V. Frouin, “Multi-object deform-able templates dedicated to the segmentation of brain deep structures.,” in W. M. Wells and A. C. F. Colchester, Proc. MICCAI, vol. 1496 of Lecture Notes in Computer Science, pp. 1134-1143, Springer Verlag, Berlin,1998.

[173] B. P. F. Lelieveldt, J. T. Rijsdam, R. J. van der Geest, D. P. Huijsmans, and J. H. C. Reiber, “Model driven interpretation of velocity encoded aortic flow images by means of Voronoi Arrangement Matrices.,” Proc. Computers in Cardiology 1998, vol. 25, pp. 753-756, 1998.

[174] M. Sonka, S. K. Tadikonda, and S. M. Collins, “Knowledge-based interpretation of MR brain images,” IEEE Transactions on Medical Imaging, vol. 15(4), pp. 443-452, 1996.

[175] S. Dellepiane, C. Regazzoni, S. B. Serpice, and G. Vernazza, “Extension of IBIS for 3D organ recog-nition in NMR multislices,” Pattern Recogrecog-nition Letters, vol. 8, pp. 65-72, 1988.

[176] H. Niemann, G. F. Sagerer, S. Schröder, and F. Kummert, “ERNEST: A semantic network system for pattern recognition,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 12(9), pp. 883-905, 1990.

[177] M. Sonka, W. Y. Park, and E. A. Hoffman, “Rule based detection of intrathoracic airway trees,” IEEE

Transactions on Medical Imaging, vol. 15(3), pp. 314-326., 1996.

(34)

3

Model driven interpretation of

velocity encoded aortic flow

images by means of Voronoi

arrangements

B P F Lelieveldt, J T Rijsdam, R J van der Geest, D P Huijsmans, J H C Reiber In: IEEE Computers in Cardiology, vol. 25, pp. 753-756, 1998.

Abstract

In this paper, we investigate the application of a representation for tomographic similar-ity to recognize vessels in aortic flow images. Due to the strict acquisition protocol, the spatial embedding of different vessels in aortic flow images is largely similar for different subjects. By modeling the vessel embedding of these images by means of Voronoi arrangements, the spatial arrangement of a ‘training’ image can be utilized to label unknown objects in target images by matching their arrangements.

The arrangement matching method was tested on routinely acquired flow images of 12 subjects, in which four major vessels were drawn manually. In all cases, the arrange-ment matching labeled the vessels in the images correctly, though clear anatomical differ-ences in size and shape were present. This shows that the arrangement metric is a powerful representation of the content of images, in which objects differ in size, shape and location, but of which the spatial embedding is similar.

3.1 Introduction

MR flow velocity mapping has proven to be a valuable modality to quantitate flow vol-ume and velocity in the Aorta. To assess left ventricular function from these images, it is necessary to outline the contours of the Aorta in a time-series of images. Previously, methods have been reported to semi-automatically trace the Aorta in a time series of Aortic flow images [1]. The purpose of this study was to develop tools to further auto-mate this procedure by utilizing prior knowledge about the spatial embedding of differ-ent vessels as they appear in Aortic flow images.

(35)

as is illustrated in Figure 3.1. The spatial configuration of the larger vessels in these images shows great similarities, although the shapes, sizes and positions of the individual vessels vary.

In this paper we investigate the applicability of a description for this tomographic similarity, which captures the spatial topology of a set of objects in an image in the so-called Voronoi arrangement. The concept of arrangement was introduced by Tagare [2, 3] as an elegant representation for the pictorial content of tomographic cross-sections, which has a specific advantage over e.g. region adjacency graphs (rotational invariance). It describes the image content through an intermediate image representation: the Voronoi diagram, which has the attractive property that for two images containing the same objects, in which the objects may differ in size and shape, the Voronoi diagrams of these images are topologically similar. Tagare’s application for arrangements is directed to image retrieval of tomographic images from an image database based on an example image; the image content of the images in the database is expressed in Voronoi arrange-ments. Based on an example image, the images in the database are selected with similar arrangements. In this paper we propose a new application for the Voronoi arrangement metric: the labelling of unknown objects in an image based on an example arrangement.

The paper is structured as follows. Section 3.2 introduces the background of Voronoi diagrams and arrangements. Section 3.3 describes a method to map an example arrange-ment matrix on a target image arrangearrange-ment, of which the labels are unknown. Section 3.4 presents a pilot validation study in which we test the arrangement labeling method on a set of routinely acquired aortic flow images from 12 different subjects. Section 3.5 concludes with a discussion.

Anatomical models in cardiovascular image analysis

Anatomical models in cardiovascular

image analysis

Contents

1

Introduction

2

Anatomical models in medical

image analysis

3

Model driven interpretation of

velocity encoded aortic flow

images by means of Voronoi

arrangements