Showing papers by "Marek Belohlavek published in 1996"

Image Analysis Using Modified Self-Organizing Maps: Automated Delineation of the Left Ventricular Cavity Boundary in Serial Echocardiograms

Abstract: An algorithm is described for delineation and three-dimensional (3D) reconstruction of the left ventricular (LV) cavity boundary from echocardiographic images The algorithm combines advanced image analysis and neural network techniques with a priori knowledge about LV shapes Minimal user interaction is required to initiate the process which results in computer-generated outlines of the LV Laboratory tests with dog hearts compare the algorithm to a conventional method based on endocardial outlines by an expert LV volume and shape comparisons are assessed

Extraction of Endocardial Boundary From Echocardiographic Images by Means of the Kohonen Self-Organizing Map

Abstract: There is considerable clinical interest in development of algorithms for reproducible determination of endocardial boundaries in echocardiographic images.1-9 This is feasible in the era of computer-aided analysis of cardiac morphology and function. However, ultrasound images are notoriously difficult to process because they are typically incomplete (dropouts, noise, etc.). Thus, automatic endocardial detection techniques require image enhancement to deal with discontinuous border definition. Our initial experiences with self-organizing maps (SOM) for the delineation of endocardial echoes is very encouraging and discussed in this manuscript. The objective was to determine whether this form of neural network combined with algorithms for edge detection can perform reproducible automated endocardial boundary delineation in artifact-prone echocardiographic images. The SOM has been preferred because: 1) no external operator is necessary to oversee the learning process of the unsupervised neural net, 2) it can be initialized with certain target-relevant shapes, 3) topological relationships are maintained between the neural net lattice nodes (the nodes define the vertices of surface tiles which may be useful for curved distances, surface area, and volume calculation), and 4) similar SOM algorithms have been successfully applied to other complex images.10