Catálogo de publicaciones - libros

In this paper we propose a Particle Filter-based approach for the segmentation of coronary arteries. To this end, successive planes of the vessel are modeled as unknown states of a sequential process. Such states consist of the orientation, position, shape model and appearance (in statistical terms) of the vessel that are recovered in an incremental fashion, using a sequential Bayesian filter (Particle Filter). In order to account for bifurcations and branchings, we consider a Monte Carlo sampling rule that propagates in parallel multiple hypotheses. Promising results on the segmentation of coronary arteries demonstrate the potential of the proposed approach.

Associating specific gene activity with functional locations in the brain results in a greater understanding of the role of the gene. To perform such an association for the over 20,000 genes in the mammalian genome, reliable automated methods that characterize the distribution of gene expression in relation to a standard anatomical model are required. In this work, we propose a new automatic method that results in the segmentation of gene expression images into distinct anatomical regions in which the expression can be quantified and compared with other images. Our method utilizes shape models from training images, texture differentiation at region boundaries, and features of anatomical landmarks, to deform a subdivision mesh-based atlas to fit gene expression images. The subdivision mesh provides a common coordinate system for internal brain data through which gene expression patterns can be compared across images. The automated large-scale annotation will help scientists interpret gene expression patterns at cellular resolution more efficiently.

In this paper, we focus on automatic kidneys detection in 2D abdominal computed tomography (CT) images. Identifying abdominal organs is one of the essential steps for visualization and for providing assistance in teaching, clinical training and diagnosis. It is also a key step in medical image retrieval application. However, due to gray levels similarities of adjacent organs, contrast media effect and relatively high variation of organ’s positions and shapes, automatically identifying abdominal organs has always been a challenging task. In this paper, we present an original method, in a statistical framework, for fully automatic kidneys detection. It makes use of spatial and gray-levels prior models built using a set of training images. The method is tested on over 400 clinically acquired images and very promising results are obtained.

This paper presents a novel approach for image segmentation by introducing competition between neighboring shape models. Our method is motivated by the observation that evolving neighboring contours should avoid overlapping with each other and this should be able to aid in multiple neighboring objects segmentation. A novel energy functional is proposed, which incorporates both prior shape information and interactions between deformable models. Accordingly, we also propose an extended maximum (MAP) shape estimation model to obtain the shape estimate of the organ. The contours evolve under the influence of image information, their own shape priors and neighboring MAP shape estimations using level set methods to recover organ shapes. Promising results and comparisons from experiments on both synthetic data and medical imagery demonstrate the potential of our approach.

Conventional level set based image segmentations are performed upon certain underlying grid/mesh structures for explicit spatial discretization of the problem and evolution domains. Such computational grids, however, lead to typically expensive and difficult grid refinement/remeshing problems whenever tradeoffs between time and precision are deemed necessary. In this paper, we present the idea of performing level set evolution in a point-based environment where the sampling location and density of the domains are adaptively determined by level set geometry and image information, thus rid of the needs for computational grids and the associated refinements. We have implemented the general geometric deformable models using this representation and computational strategy, including the incorporation of region-based prior information in both domain sampling and curve evolution processes, and have evaluated the performance of the method on synthetic data with ground truth and performed surface segmentation of brain structures from three-dimensional magnetic resonance images.

A new variational PDE based level set method for a simultaneous image segmentation and non-rigid registration using prior shape and intensity information is presented. The segmentation is obtained by finding a non-rigid registration to the prior shape. The non-rigid registration consists of both a global rigid transformation and a local non-rigid deformation. In this model, a prior shape is used as an initial contour which leads to decrease the numerical calculation time. The model is tested against two chamber end systolic ultrasound images from thirteen human patients. The experimental results provide preliminary evidence of the effectiveness of the model in detecting the boundaries of the incompletely resolved objects which were plagued by noise, dropout, and artifact.

The cerebral cortex is composed of regions with distinct laminar structure. Functional neuroimaging results are often reported with respect to these regions, usually by means of a brain “atlas”. Motivated by the need for more precise atlases, and the lack of model-based approaches in prior work in the field, this paper introduces a novel approach to parcellating the cortex into regions of distinct laminar structure, based on the theory of target tracking. The cortical layers are modelled by hidden Markov models and are tracked to determine the Bayesian evidence of layer hypotheses. This model-based parcellation method, evaluated here on a set of histological images of the cortex, is extensible to 3-D images.

The Large Scale Digital Cell Analysis System (LSDCAS) developed at the University of Iowa provides capabilities for extended-time live cell image acquisition. This paper presents a new approach to quantitative analysis of live cell image data. By using time as an extra dimension, level set methods are employed to determine cell trajectories from 2D + time data sets. When identifying the cell trajectories, cell cluster separation and mitotic cell detection steps are performed. Each of the trajectories corresponds to the motion pattern of an individual cell in the data set. At each time frame, number of cells, cell locations, cell borders, cell areas, and cell states are determined and recorded. The proposed method can help solving cell analysis problems of general importance including cell pedigree analysis and cell tracking. The developed method was tested on cancer cell image sequences and its performance compared with manually-defined ground truth. The similarity Kappa Index is 0.84 for segmentation area and the signed border positioning segmentation error is 1.6 ± 2.1 m.

We present a statistical framework that combines the registration of an atlas with the segmentation of magnetic resonance images. We use an Expectation Maximization-based algorithm to find a solution within the model, which simultaneously estimates image inhomogeneities, anatomical labelmap, and a mapping from the atlas to the image space. An example of the approach is given for a brain structure-dependent affine mapping approach. The algorithm produces high quality segmentations for brain tissues as well as their substructures. We demonstrate the approach on a set of 22 magnetic resonance images. In addition, we show that the approach performs better than similar methods which separate the registration and segmentation problems.

Intravascular ultrasound (IVUS) produces images of arteries that show the lumen in addition to the layered structure of the vessel wall. A new automatic 3D IVUS fast-marching segmentation model is presented. The method is based on a combination of region and contour information, namely the gray level probability density functions (PDFs) of the vessel structures and the image gradient. Accurate results were obtained on in-vivo and simulated data with average point to point distances between detected vessel wall boundaries and validation contours below 0.105 mm. Moreover, Hausdorff distances (that represent the worst point to point variations) resulted in values below 0.344 mm, which demonstrate the potential of combining region and contour information in a fast-marching scheme for 3D automatic IVUS image processing.