Fetal left atrium segmentation using Kohonen maps to measure the septum primum redundancy index

Abstract

Summary form only given. Echocardiographic images are used by physicians in early detection of congenital heart diseases. Ultrasonic imaging has been the basis of noninvasive methods for early detection of fetal heart diseases. However, echocardiographic images are contaminated by speckle noise, and other imaging disturbances, making it difficult to visualize important heart structures. Usually the diagnosis is obtained by measurements on the echocardiographic images. One important measure is the redundancy index of the septum primum that is associated with premature atrial contractions and the thickness of septum interventricular that can indicate the presence of myocardial hypertrophy in the fetus. The redundancy index of septum primum was obtained by ratio ledger between the maximum excursion of the septum primum (SP) to inside of left atrium and the maximum diameter of left atrium, both during diastole. For images of fetal echocardiography exams, we use Kohonen self-organizing maps (SOM) to segment and afterwards obtain measures that can help the physicians in the analysis of several congenital cardiopathies. The SOM organizes unknown data into groups of similar patterns, according to a similarity criterion. An important feature of this neural network is its ability to process noisy data. For this reason, the SOM approach has been recommended to process echocardiographic images. In this work, random samples of gray tones means of the images were used to train the map.