New insights in echocardiography based left-ventricle dynamics assessment

Abstract

Cardiovascular diseases affect a high percentage of people worldwide, being currently a major clinical concern. Echocardiograms are useful exams that allow monitoring the heart dynamics. However, their analysis depends on trained physicians with well-developed skills to recognize pathology from morphological and dynamical cues. Furthermore, these exams are often difficult to interpret due to image quality. Therefore, automatic systems able to analyze echocardiographic quantitative parameters in order to convey useful information will provide a great help in clinical diagnosis. A robust dataset was built, comprising variables associated with left-ventricle dynamics, which were studied in order to build a classifier able to discriminate between pathological and non-pathological records. To accomplish this goal, a network classifier based on decision tree was developed, using as input the left ventricle velocity over a complete cardiac cycle. This classifier revealed both sensitivity and specificity over 90% in discriminating non-pathological records, or pathological records (dilated or hypertrophic).