Detecting anomalies from echocardiography using multi-view regression of clinical measurements
Automated methods for detection of abnormalities in echocardiography are valuable for screening and routing patients with various cardiovascular diseases. Existing methods of abnormality detection are specific to the disease and usually involve extraction of various anatomical structures seen in an echocardiogram and hence, are dependent on accurate segmentation methods. In this paper, we propose a novel method of determining abnormalities from echocardiograms by directly regressing for key measurements using suitable deep learning networks. This method is illustrated in the automatic discrimination of normal and dilated cardiomyopathy cases. Specifically, a network is trained on the association of specific echocardiogram views with clinical measurements taken from such views. We show the superiority of this method of assessing abnormality by comparing it to the conventional region extraction and region characterization methods.