View Classification of Color Doppler Echocardiography via Automatic Alignment Between Doppler and B-Mode Imaging

Abstract

Echocardiography serves as a gold standard for diagnostic imaging in cardiovascular disease since it is non-intrusive, minimally invasive, and affordable. Recent advancements in deep learning techniques allowed the practical application of computer-assisted echocardiography imaging analysis, such as view classification, image segmentation, and disease diagnosis. However, unlike the more commonly investigated brightness (B-mode) imaging, there is limited research and open-source tools for the automatic processing of color Doppler echocardiography imaging (CDI) due to its more specific application and more heterogeneous image features (color flow overlaid on brightness images). Thus in this work, we developed a general framework to perform view classification of the Doppler echocardiography by leveraging the existing view classification algorithms (e.g., EchoCV) on B-mode imaging. Specifically, we developed a deep feature embedding-based module to automatically align CDI and B-mode videos based on the distance between their low-dimensional embedding. The proposed framework was evaluated on a dataset consisting of 250 subjects with ground-truth view labels by human annotators.