URL:

http://dx.doi.org/10.1109/CNNA.2006.341646

Keywords:

own

Abstract:

A novel experimental system is introduced with the ultimate aim of providing a semi-automated tool for clinicians to quantify cardiac function of the left ventricle (LV) using two-chamber transthoracic 2D video flows. The algorithm extracts space-time signatures from echocardiograms based on real-time pixel-level parallel boundary tracking methods (topographic cellular active contour techniques) that are used to drive learning and recognition modules in order to enhance reliability of diagnoses made up by clinicians. In extraction process subsequent ellipse model based wall-segment identification, sampling and filtering are used, furthermore specific wall segment thickening models for region of interest (ROI) selection and reliability enhancement of segment identification and signature interpretation. Simple learning algorithms like decision tree and fuzzy produced adequate results in the first stage of experiments on large database of patient records