Quantification of cardiac bull's-eye map based on principal strain analysis for myocardial wall motion assessment in stress echocardiography

In this paper we consider automated myocardial wall motion assessment by quantifying a cardiac bull's eye map derived from principal strain analysis. The objective is to learn a classification model that can classify between normal and abnormal wall motions. A traditional hand-crafted feature approach based on pixel intensities is compared with a deep learning framework, where a Convolutional Neural Network (CNN) automatically learns features. Experiments on a 3D Dobutamine Stress Echo (DSE) dataset with normal and abnormal wall motions shows that both hand-crafted approaches yield comparable accuracy: Random Forests (72.1%), Support Vector Machines (70.5%), and CNN at a slightly higher accuracy (75.0%) and a lower training computational cost.