Catálogo de publicaciones - libros

Multi–slice computed tomography image series are a valuable source of information to extract shape and motion parameters of the heart. We present a method how to segment and label all main chambers (both ventricles and atria) and connected vessels (arteries and main vein trunks) from such images and to track their movement over the cardiac cycle. A framework is presented to construct a multi–surface triangular model enclosing all blood–filled cavities and the main myocardium as well as to adapt this model to unseen images, and to propagate it from phase to phase. While model construction still requires a reasonable amount of user interaction, adaptation is mostly automated, and propagation works fully automatically. The adaptation method by deformable surface models requires a set of landmarks to be manually located for one of the cardiac phases for model initialisation.

The relationships among vascular geometry, hemodynamics, and plaque development in coronary arteries are not yet well understood. This in-vivo study was based on the observation that plaque frequently develops at the inner curvature of a vessel, presumably due to a relatively lower wall shear stress. We have shown that circumferential plaque distribution depends on the vessel curvature in the majority of vessels. Consequently, we studied the correlation of plaque distribution and hemodynamics in a set of 48 vessel segments reconstructed by 3-D fusion of intravascular ultrasound and x-ray angiography. The inverse relationship between local wall shear stress and plaque thickness was significantly more pronounced (<0.025) in vessel cross sections exhibiting compensatory enlargement (positive remodeling) without luminal narrowing than when the full spectrum of vessel stenosis severity was considered. Our findings confirmed that relatively lower wall shear stress is associated with increased plaque development.

A novel approach to automated segmentation of X-ray Left Ventricu-lar (LV) angiograms is proposed, based on Active Appearance Models (AAMs) and dynamic programming (DP). Due to combined modeling of the end-diastolic (ED) and end-systolic (ES) phase, existing correlations in shape and texture representation are exploited, resulting in a better segmentation in the ES phase. The intrinsic over-constraining by the model is compensated by a DP algorithm, in which also cardiac contraction motion features are incorporated. An elaborate evaluation of the algorithm, based on 70 paired ED-ES images, shows success rates of 100% for ED and 99% for ES, with average border positioning errors of 0.68 mm and 1.45 mm respectively. Calculated volumes were accurate and unbiased, proving the high clinical potential of our method.

In this paper, a new technique (SPASM) based on a 3D-ASM is presented for automatic segmentation of cardiac MRI image data sets consisting of multiple planes with different orientations, and with large undersampled regions. SPASM was applied to sparsely sampled and radially oriented cardiac LV image data.

Performance of SPASM has been compared to results from other methods reported in literature. The accuracy of SPASM is comparable to these other methods, but SPASM uses considerably less image data.

A geometrical model of the human heart is of interest in many fields of biophysics. The myocardium contains the electrical sources responsible for the generation of the body-surface ECG. An accurate geometric knowledge of these sources is crucial when dealing with the electrocardiographic forward and inverse problem. We developed a semiautomatic approach for segmenting the myocardium in order to deal with the electrocardiographic problem. The approach can be divided into two main steps. The first step extracts the atrial and ventricular blood masses by employing Active Appearance Models (AAM). The ventricular blood masses are segmented automatically after providing the positions of the apex cordis and the base of the heart. Due to the complex geometry of the atria the segmentation process of the atrial blood masses requires more information. We divided, therefore, the left and the right atrium into three divisions of appearance: the base of the heart, the lower pulmonary veins from its first up to the last appearance in the image stack, and the upper pulmonary veins. After successful extraction of the blood masses the second step involves morphologically-based operations in order to extract the myocardium either directly by detecting the myocardium in the volume block, or by reconstructing the myocardium using mean model information, in case the algorithm fails to detect the myocardium.

This paper describes a method for automatic contour detection in long-axis cardiac MRI using an adaptive virtual exploring robot. The robot is a simulated trained virtual autonomous tri-cycle that is initially positioned in a binary representation of the left ventricle (LV) and finds the contours during navigation through the ventricle. The method incorporates global and local prior shape knowledge of the LV in order to adapt the navigational parameters. Together with kinematic constraints, the robot is able to avoid concave regions such as papillary muscles and navigate through narrow corridors such as the apex. Validation was performed on in-vivo multiphase long-axis cardiac MRI images of 11 subjects. Results showed good correlation between the quantitative parameters, computed from manual and automatic segmentation: for end-diastolic volume (EDV) r=0.91, for end-systolic volume (ESV) r=0.93, ejection fraction (EF) r=0.77, and LV mass (LVM) r=0.80.

Plaque analysis in IVUS planes needs accurate intima and adventitia models. Large variety in adventitia descriptors difficulties its detection and motivates using a classification strategy for selecting points on the structure. Whatever the set of descriptors used, the selection stage suffers from fake responses due to noise and uncompleted true curves. In order to smooth background noise while strengthening responses, we apply a restricted anisotropic filter that homogenizes grey levels along the image significant structures. Candidate points are extracted by means of a simple semi supervised adaptive classification of the filtered image response to edge and calcium detectors. The final model is obtained by interpolating the former line segments with an anisotropic contour closing technique based on functional extension principles.

A procedure for helping the professional in electrophysiology in performing catheter ablation as a definitive treatment of certain types of arrythmia is presented here. This procedure uses trajectory planning techniques that have been developed in the robotics field. Starting off from signals obtained in an electrophysiological study of a patient, an electrical model of the heart with zones of different propagation properties is generated. Trajectory planning techniques are used to obtain the qualitative behavior of the heart under different types of arrythmia. A good point for ablation is computed as one that interrupts the trajectory that is sustaining the arrythmia.

We developed a 3D computer graphic model of functional anatomy of the human heart. The model provides visually correct anatomical and functional detail suitable for medical education. We reconstructed 3D surface models of the human heart based on segmentation obtained from the Visible Human image datasets. We developed a fiber based muscle action model specially adapted for the myocardium. Each muscle fiber is equipped with contractile and elastic elements and is used as a local shape deformation guide. The timing of fiber contraction activation is driven by patient specific action potential excitation patterns. As a first step we have visualized the function of a healthy heart. We are now planning to visualize a range of cardiac conditions and dysfunctions.

Different methods were evaluated to enlarge artificially a training set which is used to build a statistical shape model. In this work, the shape model was built from MR data of 25 subjects and it consisted of ventricles, atria and epicardium. The method adding smooth non-rigid deformations to original training set examples produced the best results. The results indicated also that artificial deformation modes model better an unseen object than an equal number of standard PCA modes generated from original data.