Cardiac Anatomy-ware Echocardiography Wall Motion Estimation

Medical ultrasound is extensively used in diagnosing cardiovascular diseases, offering real-time visualization of the beating heart and enabling the assessment of complex cardiac wall functionality. Cardiac wall motion is a crucial biomarker for diagnosing heart failure and myocardial infarction. In general, speckle tracking and optical flow estimation have been the primary methods for quantifying cardiac wall motion by analyzing pixel similarities in myocardial regions across consecutive image frames. However, the methods have limitations, as they lack a semantic understanding of cardiac anatomy and rely solely on the pixel attributes of echocardiography images. The dependency results in significant performance degradation, particularly when image quality is poor or portions of the cardiac wall are obscured. In this paper, we introduce a cardiac anatomy-aware optical flow estimation neural network that incorporates a semantic understanding of cardiac wall anatomy, thereby enhancing the accuracy of cardiac wall motion estimation.