Catálogo de publicaciones - libros

Shape matching plays a prominent role in the analysis of medical and biological structures. Recently, a unifying framework was introduced for shape matching that uses mixture-models to couple both the shape representation and deformation. Essentially, shape distances were defined as geodesics induced by the Fisher-Rao metric on the manifold of mixture-model represented shapes. A fundamental drawback of the Fisher-Rao metric is that it is NOT available in closed-form for the mixture model. Consequently, shape comparisons are computationally very expensive. Here, we propose a new Riemannian metric based on generalized - entropy measures. In sharp contrast to the Fisher-Rao metric, our new metric is available in closed-form. Geodesic computations using the new metric are considerably more efficient. Discriminative capabilities of this new metric are studied by pairwise matching of corpus callosum shapes. Comparisons are conducted with the Fisher-Rao metric and the thin-plate spline bending energy.

Patient specific modelling of the blood flow through the human aorta is performed using computational fluid dynamics (CFD) and magnetic resonance imaging (MRI). Velocity patterns are compared between computer simulations and measurements. The workflow includes several steps: MRI measurement to obtain both geometry and velocity, an automatic levelset segmentation followed by meshing of the geometrical model and CFD setup to perform the simulations follwed by the actual simulations. The computational results agree well with the measured data.

In this study an approach for testing electrode array schemes with respect to their ability to improve the resolution of methods for activation time imaging is proposed. First local linear dependency maps are computed using a virtual array method. These maps depict the torso surface areas where the body surface potential is most sensitive to changes in the transmembrane potential. The optimal number and position of the electrodes within the sensitive body surface regions was selected by constructing effort gain (EG) plots. Such a plot depicts the relative attainable rank of the leadfield matrix in relation to the increase in number of electrodes required to build the electrode array.

From the sensitivity maps it was found that the BSP is most sensitive to changes in TMP on the upper left frontal and dorsal body surface. The EG analysis revealed that the optimal array meeting clinical requirements and improving the resolution of activation time imaging consists of 125 electrodes.

Automatic algorithms in conjunction with longitudinal MR brain images can be used to measure cerebral atrophy, which is particularly pronounced in several types of dementia. An atrophy simulation technique has been devised to facilitate validation of these algorithms. To make this model of atrophy more realistic we simulate acquisition artefacts which are common problems in dementia imaging: motion (both step and periodic motion) and pulsatile flow artefact. Artefacts were simulated by combining different portions of k-space from various modified image. The original images were 7 MR scans of healthy elderly controls, each of which had two levels of simulated atrophy. We investigate the effect of the simulated acquisition artefacts in atrophy measurements provided by an automatic technique, SIENA.

This paper investigates the use of Active Shape Models (ASM) to capture the variability of the intra-thoracic airway tree. The method significantly reduces the dimensionality of the non-rigid 2D/3D registration problem and leads to a rapid and robust registration framework. In this study, EM tracking data has been also incorporated through a probabilistic framework for providing a statistically optimal pose given both the EM and the image-based registration measurements. Comprehensive phantom experiments have been conducted to assess the key numerical factors involved in using catheter tip EM tracking for deformable 2D/3D registration.

This paper presents a novel approach for multi-organ (musculoskeletal system) automatic registration and segmentation from clinical MRI datasets, based on discrete deformable models (simplex meshes). We reduce the computational complexity using multi-resolution forces, multi-resolution hierarchical collision handling and large simulation time steps (implicit integration scheme), allowing real-time user control and cost-efficient segmentation. Radial forces and topological constraints (attachments) are applied to regularize the segmentation process. Based on a medial axis constrained approximation, we efficiently characterize shapes and deformations. We validate our methods for the hip joint and the thigh (20 muscles, 4 bones) on 4 datasets: average error=1.5mm, computation time=15min.

We propose here a framework to build a statistical atlas of diffusion tensors of canine hearts. The anatomical images of seven hearts are first non-rigidly registered in the same reference frame and their associated diffusion tensors are then transformed with a method that preserves the cardiac laminar sheets. In this referential frame, the mean tensor and its covariance matrix are computed based on the Log-Euclidean framework. With this method, we can produce a smooth mean tensor field that is suited for fiber tracking algorithms or the electromechanical modeling of the heart. In addition, by examining the covariance matrix at each voxel it is possible to assess the variability of the cardiac fiber directions and of the orientations of laminar sheets. The results show a strong coherence of the diffusion tensors and the fiber orientations among a population of seven normal canine hearts.

Modeling the deflection of flexible needles is an essential part of needle insertion simulation and path planning. In this paper, three models are compared in terms of accuracy in simulating the bending of a prostate brachytherapy needle. The first two utilize the finite element method, one using geometric non-linearity and triangular plane elements, the other using non-linear beam elements. The third model uses angular springs to model cantilever deflection. The simulations are compared with the experimental bent needle configurations. The models are assessed in terms of geometric conformity using independently identified and pre-identified model parameters. The results show that the angular spring model, which is also the simplest, simulates the needle more accurately than the others.

A motion-based Inverse Kinematics Knee (IKK) model was developed for Total Knee Replacement (TKR) joints. By tracking a sequence of passive knee motion, the IKK model estimated ligament properties such as insertion locations. The formulation of the IKK model embedded a Forward Kinematics Knee (FKK) [1] model in a numerical optimization algorithm known as the Unscented Kalman Filter [2]. Simulation results performed on a semi-constrained TKR design suggested that ligament insertions could be accurately estimated in the medial-lateral (ML) and the proximal-distal (PD) directions, but less reliably in the anterior-posterior (AP) direction for the tibial component. However, the forward kinematics produced by both the true and estimated ligament properties were nearly identical, suggesting that the IKK model recovered a set of ligament properties. These results imply that it may not be necessary to use a patient-specific CT or MRI scan to locate ligaments, which considerably widens potential applications of kinematic-based total knee replacement.

Estimation of the system parameters, given noisy input/output data, is a major field in control and signal processing. Many different estimation methods have been proposed in recent years. Among various methods, Extended Kalman Filtering (EKF) is very useful for estimating the parameters of a nonlinear and time-varying system. Moreover, it can remove the effects of noises to achieve significantly improved results. Our task here is to estimate the coefficients in a spring-beam-damper needle steering model. This kind of spring-damper model has been adopted by many researchers in studying the tissue deformation. One difficulty in using such model is to estimate the spring and damper coefficients. Here, we proposed an online parameter estimator using EKF to solve this problem. The detailed design is presented in this paper. Computer simulations and physical experiments have revealed that the simulator can estimate the parameters accurately with fast convergent speed and improve the model efficacy.