Solving the problem of calculating strain indices in echocardiography using deep learning neural networks

Abstract

The paper sets the task of calculating the parameters of deformation of the heart muscle according to echocardiogram data under interference conditions, for example, in the study of children. Indicators of deformation (strain value) of the heart muscle were used by us to determine the presence and severity of dysfunction of the chambers of the heart in atrial septal defect – a congenital heart defect characterized by the presence of communication between the right and left atria. The problem was solved by analyzing the video stream obtained from the installation of echocardiography using a set of deep learning neural network architectures designed for image segmentation. The study was conducted for the U-net architecture. As a result of processing the video stream, it was possible to solve the problem of segmentation of the walls of the heart muscle and binding of key points in the condition of interference in the removal of an echocardiogram on child patients unable to remain motionless during the study. The obtained indicators provide the cardiologist with important information for determining the dysfunction of the chambers of the heart (especially the right atrium, the most compromised chamber of the heart in the studied cases) with a defect of the atrial septum.