Catálogo de publicaciones - libros

This paper presents a method for classification of medical images, using machine learning and deformation-based morphometry. A morphological representation of the anatomy of interest is first obtained using high-dimensional template warping, from which regions that display strong correlations between morphological measurements and the classification (clinical) variable are extracted using a watershed segmentation, taking into account the regional smoothness of the correlation map which is estimated by a cross-validation strategy in order to achieve robustness to outliers. A Support Vector Machine-Recursive Feature Elimination (SVM-RFE) technique is then used to rank computed features from the extracted regions, according to their effect on the leave-one-out error bound. Finally, SVM classification is applied using the best set of features, and it is tested using leave-one-out. The results from a group of 61 brain images of female normal controls and schizophrenia patients demonstrate not only high classification accuracy (91.8%) and steep ROC curves, but also exceptional stability with respect to the number of selected features and the SVM kernel size.

We present a novel multi-scale bone enhancement measure that can be used to drive a geometric flow to segment bone structures. This measure has the essential properties to be incorporated in the computation of anatomical models for the simulation of pituitary surgery, enabling it to better account for the presence of sinus bones. We present synthetic examples that validate our approach and show a comparison between existing segmentation techniques of paranasal sinus CT data.

In this paper, we present a novel variational formulation of the registration assisted image segmentation problem which leads to solving a coupled set of nonlinear PDEs that are solved using efficient numerical schemes. Our work is a departure from earlier methods in that we have a unified variational principle wherein non-rigid registration and segmentation are simultaneously achieved; unlike previous methods of solution for this problem, our algorithm can accommodate for image pairs having very distinct intensity distributions. We present examples of performance of our algorithm on synthetic and real data sets along with quantitative accuracy estimates of the registration.

Validation and method of comparison for segmentation of magnetic resonance images (MRI) presenting pathology is a challenging task due to the lack of reliable ground truth. We propose a new method for generating synthetic multi-modal 3D brain MRI with tumor and edema, along with the ground truth. Tumor mass effect is modeled using a biomechanical model, while tumor and edema infiltration is modeled as a reaction-diffusion process that is guided by a modified diffusion tensor MRI. We propose the use of warping and geodesic interpolation on the diffusion tensors to simulate the displacement and the destruction of the white matter fibers. We also model the process where the contrast agent tends to accumulate in cortical csf regions and active tumor regions to obtain contrast enhanced T1w MR image that appear realistic. The result is simulated multi-modal MRI with ground truth available as sets of probability maps. The system will be able to generate large sets of simulation images with tumors of varying size, shape and location, and will additionally generate infiltrated and deformed healthy tissue probabilities.

Accurate automatic extraction of a 3D cerebrovascular system from images obtained by time-of-flight (TOF) or phase contrast (PC) magnetic resonance angiography (MRA) is a challenging segmentation problem due to small size objects of interest (blood vessels) in each 2D MRA slice and complex surrounding anatomical structures, e.g. fat, bones, or grey and white brain matter. We show that due to a multi-modal nature of MRA data blood vessels can be accurately separated from background in each slice by a voxel-wise classification based on precisely identified probability models of voxel intensities. To identify the models, an empirical marginal probability distribution of intensities is closely approximated with a linear combination of discrete Gaussians (LCDG) with alternate signs, and we modify the conventional Expectation-Maximization (EM) algorithm to deal with the LCDG. To validate the accuracy of our algorithm, a special 3D geometrical phantom motivated by statistical analysis of the MRA-TOF data is designed. Experiments with both the phantom and 50 real data sets confirm high accuracy of the proposed approach.

In the context of stroke therapy simulation, a method for the segmentation and reconstruction of human vasculature is presented and evaluated. Based on CTA scans, semi-automatic tools have been developed to reduce dataset noise, to segment using active contours, to extract the skeleton, to estimate the vessel radii and to reconstruct the associated surface. The robustness and accuracy of our technique are evaluated on a vascular phantom scanned in different orientations. The reconstructed surface is compared to a surface generated by marching cubes followed by decimation and smoothing. Experiments show that the proposed technique reaches a good balance in terms of smoothness, number of triangles, and distance error. The reconstructed surface is suitable for real-time simulation, interactive navigation and visualization.

Precise segmentation of three-dimensional (3D) magnetic resonance angiography (MRA) image can be a very useful computer aided diagnosis (CAD) tool in clinical routines. Our objective is to develop a specific segmentation scheme for accurately extracting vasculature from MRA images. Our proposed algorithm, called the capillary active contour (CAC), models capillary action where liquid can climb along the boundaries of thin tubes. The CAC, which is implemented based on level sets, is able to segment thin vessels and has been applied for verification on synthetic volumetric images and real 3D MRA images. Compared with other state-of-the-art MRA segmentation algorithms, our experiments show that the introduced capillary force can facilitate more accurate segmentation of blood vessels.

Graph methods that summarize vasculature by its branching topology are not sufficient for the statistical characterization of a population of intra-cranial vascular networks. Intra-cranial vascular networks are typified by topological variations and long, wandering paths between branch points.

We present a graph-based representation, called spatial graphs, that captures both the branching patterns and the spatial locations of vascular networks. Furthermore, we present companion methods that allow spatial graphs to (1) statistically characterize populations of vascular networks, (2) generate the central vascular network of a population of vascular networks, and (3) distinguish between populations of vascular networks. We evaluate spatial graphs by using them to distinguish the gender and handedness of individuals based on their intra-cranial vascular networks.

The spherical harmonic (SPHARM) description is a powerful surface modeling technique that can model arbitrarily shaped but simply connected 3D objects and has been used in many applications in medical imaging. Previous SPHARM techniques use the first order ellipsoid for establishing surface correspondence and aligning objects. However, this first order information may not be sufficient in many cases; a more general method for establishing surface correspondence would be to minimize the mean squared distance between two corresponding surfaces. In this paper, a new surface matching algorithm is proposed for 3D SPHARM models to achieve this goal. This algorithm employs a useful rotational property of spherical harmonic basis functions for a fast implementation. Applications of medical image analysis (, spatio-temporal modeling of heart shape changes) are used to demonstrate this approach. Theoretical proofs and experimental results show that our approach is an accurate and flexible surface correspondence alignment method.

We propose an algorithm for jointly performing registration point selection and interactive, rigid, surface-based registration. The registration is computed using a particle filter that outputs a sampled representation of the distribution of the registration parameters. The distribution is propagated through a point selection algorithm derived from a stiffness model of surface-based registration, allowing the selection algorithm to incorporate knowledge of the uncertainties in the registration parameters. We show that the behavior of target registration error improves as the quality measure of the registration points increases.