Edgard Nyssen

Vascular Active Contour for Vessel Tree Segmentation

In this paper, a novel active contour model is proposed for vessel tree segmentation. First, we introduce a region competition-based active contour model exploiting the Gaussian mixture model, which mainly segments thick vessels. Second, we define a vascular vector field to evolve the active contour along its center line into the thin and weak vessels. The vector field is derived from the eigenanalysis of the Hessian matrix of the image intensity in a multiscale framework. Finally, a dual curvature strategy, which uses a vesselness measure-dependent function selecting between a minimal principal curvature and a mean curvature criterion, is added to smoothen the surface of the vessel without changing its shape. The developed model is used to extract the liver and lung vessel tree as well as the coronary artery from high-resolution volumetric computed tomography images. Comparisons are made with several classical active contour models and manual extraction. The experiments show that our model is more accurate and robust than these classical models and is, therefore, more suited for automatic vessel tree extraction.

Object recognition in brain CT-scans: knowledge-based fusion of data from multiple feature extractors

Describes a knowledge-based image interpretation system for the segmentation and labeling of a series of 2-D brain X-ray CT-scans, parallel to the orbito-meatal plane. The system combines the image primitive information produced by different low level vision techniques in order to improve the reliability of the segmentation and the image interpretation. It is implemented in a blackboard environment that is holding various types of prior information and which controls the interpretation process. The scoring model is applied for the fusion of information derived from three types of image primitives (points, edges, and regions). A model, containing both analogical and propositional knowledge on the brain objects, is used to direct the interpretation process. The linguistic variables, introduced to describe the propositional features of the brain model, are defined by fuzzy membership functions. Constraint functions are applied to evaluate the plausibility of the mapping between image primitives and brain model data objects. Procedural knowledge has been integrated into different knowledge sources. Experimental results illustrate the reliability and robustness of the system against small variations in slice orientation and interpatient variability in the images.

Region competition based active contour for medical object extraction

In this paper, a probabilistic and level set model for three-dimensional medical object extraction is proposed, which is called region competition based active contour. The algorithms are derived by minimizing a region based probabilistic energy function and implemented in a level set framework. An additional speed-controlling term makes the active contour quickly convergent to the actual contour on strong edges, whereas a probabilistic model makes the active contour performing well for weak edges. Prior knowledge about the initial contour and the probabilistic distribution contributes to more efficient extraction. The developed model has been applied to a variety of medical images, from CTA and MRA of the coronary to rotationally scanned and real-time three-dimensional echocardiography images of the mitral valve. As the results show, the algorithm is fast, convergent, adapted to a broad range of medical objects and produces satisfactory results.

Markov random field-based clustering applied to the segmentation of masses in digital mammograms

In this paper we propose a new pixel clustering model applied to the analysis of digital mammograms. The clustering represents here the first step in a more general method and aims at the creation of a concise data-set (clusters) for automatic detection and classification of masses, which are typically among the first symptoms analysed in early diagnosis of breast cancer. For the purpose of this work, a set of mammographic images has been employed, that are 12-bit gray level digital scans and as such, are inherently inhomogeneous and affected by the noise resulting from the film scanning. The image pixels are described only by their intensity (gray level), therefore, the available information is limited to one dimension. We propose a Markov random field (MRF)-based technique that is suitable for performing clustering in an environment which is described by poor or limited data. The proposed method is a statistical classification model, that labels the image pixels based on the description of their statistical and contextual information. Apart from evaluating the pixel statistics, that originate from the definition of the K-means clustering scheme, the model expands the analysis by the description of the spatial dependence between pixels and their labels (context), hence leading to the reduction of the inhomogeneity of the output. Moreover, we define a probabilistic description of the model, that is characterised by a remarkable simplicity, such that its realisation can be easily and efficiently implemented in any high- or low-level programming language, thus allowing it to be run on virtually any kind of platform. Finally, we evaluate the algorithm against the classical K-means clustering routine. We point out similarities between the two methods and, moreover, show the advantages and superiority of the MRF scheme.

Using Landmarks to Establish a Point-to-Point Correspondence between Signatures

This paper introduces a new approach for point-to-point correspondence finding, which can be used as pre-processing stage of a handwritten signature verification procedure. This approach provides a solid basis for comparing function features of two handwritten signatures. Corner points of the signatures are first extracted based on velocity information. The characteristics of curvilinear velocity and angular velocity are combined successfully by functions based on membership criteria. The signatures to be compared are then segmented at landmarks obtained by corner matching based on similarity measures. In the last step, the corresponding pairs of segments are mapped by a point-to-point matching algorithm, minimising a curve deformation energy. The techniques described were applied to a set of 188 signatures from 19 volunteers. The resulting point-to-point matching of signature pairs was satisfactory in all cases where there was a visual agreement between the signatures.

Liver Segmentation by an Active Contour Model with Embedded Gaussian Mixture Model based Classifiers

Automatic liver segmentation is a crucial step for diagnosis and surgery planning. To extract the liver, its tumors and vessels, we developed an active contour model with an embedded classifier, based on a Gaussian mixture model fitted to the intensity distribution of the medical image. The difference between the maximum membership of the intensities belonging to the classes of the object and those of the background is included as an extra speed propagation term in the active contour model. An additional speed controlling term slows down the evolution of the active contour when it approaches an edge, making it quickly convergent to the ideal object. The developed model has been applied to liver segmentation. Some comparisons are made between the Geodesic Active Contour, C-V (active contour without edges) and our model. As the experiments show, our model is accurate, flexible and suited to extract objects surrounded by a complicated background.

2D image registration using focused mutual information for application in dentistry

Spatial alignment of image data is a common task in computer vision and medical imaging. This should preferentially be done with minimal intervention of an operator. Similarity measures with origin in the information theory such as mutual information (MI) have proven to be robust registration criteria for this purpose. Intra-oral radiographs can be considered images of piecewise rigid objects. Teeth and jaws are rigid but can be displaced with respect to each other. Therefore MI criteria combined with affine deformations tend to fail, when teeth and jaws move with respect to each other between image acquisitions. In this paper, we consider a focused weighing of pixels in the reference image. The resulting criterion, focused mutual information (FMI) is an adequate tool for the registration of rigid parts of a scene. We also show that the use of FMI is more robust for the subtraction of lateral radiographs of teeth, than MI confined to a region of interest. Furthermore, the criterion allows the follow-up of small carious lesions when upper and lower jaw moved between the acquisition of test and reference image.

Detection of buildings from a single airborne image using a Markov random field model

We propose an automated method for the detection of buildings from a single airborne color optical image using a dedicated Markov random field model, which describes both geometric and photometric attributes of the 3-D objects of interest. The paper presents the basic principles and some preliminary results of our approach.

Novel information theory based method for superimposition of lateral head radiographs and cone beam computed tomography images

OBJECTIVES The aim was to introduce a novel alignment criterion, focus mutual information (FMI), for the superimposition of lateral cephalometric radiographs and three dimensional (3D) cone beam computed images as well as the assessment of the alignment characteristics of the new method and comparison of the novel methodology with the region of interest (ROI) approach. METHODS Implementation of a FMI criterion-based methodology that only requires the approximate indication of stable structures in one single image. The robustness of the method was first addressed in a phantom experiment comparing the new technique with a ROI approach. Two consecutive cephalometric radiographs were then obtained, one before and one after functional twin block application. These images were then superimposed using alignment by FMI where the following were focused on, in several ways: (1) cranial base and acoustic meatus, (2) palatal plane and (3) mandibular symphysis. The superimposed images were subtracted and coloured. The applicability to cone beam CT (CBCT) is illustrated by the alignment of CBCT images acquired before and after craniofacial surgery. RESULTS The phantom experiment clearly shows superior alignment when compared to the ROI approach (Wilcoxon n = 17, Z = -3.290, and P = 0.001), and robustness with respect to the choice of parameters (one-sample t-test n = 50, t = -12.355, and P = 0.000). The treatment effects are revealed clearly in the subtraction image of well-aligned cephalometric radiographs. The colouring scheme of the subtraction image emphasises the areas of change and visualizes the remodelling of the soft tissue. CONCLUSIONS FMI allows for cephalometry without tracing, it avoids the error inherent to the use of landmarks and the interaction of the practitioner is kept to a minimum. The robustness to focal distribution variations limits the influence of possible examiner inaccuracy.

Statistical models for multidisciplinary applications of image segmentation and labelling

Three classes of statistical techniques used to solve image segmentation and labelling problems are reviewed: (1) supervised and unsupervised pixel classification, (2) exploitation of the probability distribution map as a way to model image structure, (3) Markov random field modelling combined with MAP statistical classification. Diverse examples illustrate the potential of the three approaches that are described as generic methods belonging to a common framework for image segmentation/labelling.